Field Sobriety Tests

Field sobriety test accuracy figures like “91 percent” come from a handful of NHTSA-funded studies done between 1977 and 1998. Those studies measured something narrower than most people think: how often officers who used the tests correctly guessed whether a driver was above or below a set alcohol level. They did not measure whether a person was impaired, and they were not designed to prove anyone’s exact BAC.

76%officer decisions correct, 1977 lab study (0.10 criterion)
81%volunteers correctly classified, 1981 lab study
91%arrest decisions correct, 1998 San Diego (0.08)
0.122average measured BAC of the San Diego drivers

If you were arrested for DWI in North Carolina, the officer probably asked you to follow a pen with your eyes, walk a line heel to toe, and stand on one leg. Those three exercises are the Standardized Field Sobriety Tests (SFSTs), and in court they are often described as “scientifically validated.” This guide walks through each of the original validation studies, the actual numbers from the primary NHTSA reports, what critics have found, and what it all means under North Carolina law. We have tried to report every percentage exactly as the source documents state it.

What does “field sobriety test accuracy” actually mean?

In the NHTSA research, “accuracy” means the share of officer decisions (arrest or release) that matched a chemical test, measured against one alcohol cutoff. It is a classification score, not a measure of driving impairment.

Every study used the same basic design. Officers gave the tests, wrote down their scores and their decision, and researchers then compared that decision with a breath or blood result. If the officer said “over” and the chemical test was over the study’s cutoff, that counted as a correct decision. The cutoff changed from study to study: 0.10 in 1977 and 1981, 0.05 in Colorado, and 0.08 in Florida and San Diego. That matters, because a percentage earned at one cutoff says nothing direct about another.

It also matters for North Carolina, where the State can prove DWI two main ways under G.S. 20-138.1: appreciable impairment, or an alcohol concentration of 0.08 or more. The SFST studies were built to predict the second, a number, yet in court the tests are usually offered to prove the first.

Black-and-white 1973 photo of a police chief inspector demonstrating a field sobriety test during a roadside check
A 1973 roadside sobriety check, from the era before the tests were standardized. Photo: Jukka Kuusisto, Wikimedia Commons, CC BY 4.0.

The 1977 Burns and Moskowitz lab study: where the battery began

The 1977 study found that officers using a mix of six roadside tests made the correct over-or-under-0.10 decision 76 percent of the time, and that nearly half of their arrest decisions were wrong.

NHTSA hired the Southern California Research Institute to find the best roadside tests. In Psychophysical Tests for DWI Arrest (DOT HS-802 424, June 1977), Marcelline Burns and Herbert Moskowitz had 10 police, sheriff and highway patrol officers test 238 volunteers who had been given alcohol in a laboratory. The officers tried six tests: one-leg stand, walk-and-turn, finger-to-nose, finger count, “alcohol gaze nystagmus” and tracing.

Officer decisions were correct 76 percent of the time. The less-quoted number is the error rate among arrests: 47 of the 101 people the officers would have arrested had BACs below 0.10. The researchers recommended keeping three tests: the one-leg stand, the walk-and-turn and gaze nystagmus.

A volunteer attempts a finger-to-nose test during a simulated field sobriety test given by a sheriff's deputy
Finger-to-nose was one of the six tests studied in 1977. It was dropped from the standardized battery. Photo: U.S. Navy photo by Mass Communication Specialist 2nd Class Charles White, Wikimedia Commons, Public domain.

The 1981 study: standardizing the three tests

The 1981 report standardized HGN, walk-and-turn and one-leg stand and found that test scores correctly sorted 81 percent of lab volunteers as above or below 0.10.

In Development and Field Test of Psychophysical Tests for DWI Arrest (DOT HS-805 864, March 1981), V. Tharp, Burns and Moskowitz wrote the standardized instructions and scoring that officers still learn. Ten officers tested 297 drinking volunteers with BACs from 0 to 0.18. Scores classified 81 percent correctly against 0.10, and officers’ BAC estimates were off by about 0.03 on average. Reliability between scorers and on retesting ranged from 0.60 to 0.80. A limited field phase gathered 3,128 data forms over three months.

NHTSA’s training materials later broke the 1981 results down by test. As quoted in the federal court decision discussed below, HGN alone was about 77 percent accurate, walk-and-turn 68 percent and one-leg stand 65 percent, with HGN and walk-and-turn combined at 80 percent. Psychologists who later reviewed the work pointed out that 32 percent of the lab subjects judged to be at 0.10 or above were actually below it (Cole & Nowaczyk, 1994).

  1. 1977Burns & Moskowitz lab study: 238 volunteers, six tests, 0.10 cutoff, 76% of decisions correct.
  2. 1981Tharp, Burns & Moskowitz: three-test battery standardized; 81% correct classification in the lab.
  3. 1983NHTSA field evaluation (DOT HS-806 475), with data from Maryland, Virginia, North Carolina and D.C. according to a CHP summary.
  4. 1995Colorado field study: 234 drivers with known BACs, 0.05 cutoff, 86% correct.
  5. 1997Florida (Pinellas County) field study: 0.08 cutoff, 197 of 206 arrests correct.
  6. 1998San Diego field study: 297 drivers, 0.08 cutoff, 91% correct.

The field validation studies: Colorado, Florida and San Diego

The three 1990s field studies reported higher numbers, 86 to 95 percent depending on how you count, but each used a different cutoff and each studied drivers who had already been stopped and suspected.

Colorado, 1995

Marcelline Burns and Deputy Ellen Anderson of the Pitkin County Sheriff’s Office collected 305 records from volunteer Colorado agencies (final report, November 1995). Usable BACs existed for 234. Using 0.05 as the line, officers were right in 86 percent of those cases (201 of 234). Arrests were 93 percent correct, but releases were only 64 percent correct. Officers did not carry portable breath testers during the study.

Florida, 1997

Burns and Sergeant Teresa Dioquino ran the Florida study with Pinellas County deputies from June to September 1997, using Florida’s 0.08 limit. Of 379 contacts, 256 had a measured BAC. Deputies’ arrests were correct in 197 of 206 cases (about 95 percent), and releases were correct in 41 of 50 (82 percent). Some released drivers’ BACs came from roadside screening devices run by observers rather than evidential instruments.

San Diego, 1998

Jack Stuster and Burns trained seven officers from the San Diego Police Department’s alcohol enforcement unit (DOT HS 808 839, August 1998). Across 297 motorists, arrest-or-release decisions were correct 91 percent of the time at 0.08. By test, HGN was 88 percent accurate, one-leg stand 83 percent and walk-and-turn 79 percent. Scoring was adjusted so that four HGN clues pointed to 0.08 or more and two clues to 0.04 or more. Officers arrested 217 of the 297 drivers, and the average measured BAC was 0.122.

Reported accuracy of officer decisions, by study

1977 lab (cutoff 0.10)76%
1981 lab (cutoff 0.10)81%
1995 Colorado (cutoff 0.05)86%
1997 Florida, arrests only (0.08)~95%
1998 San Diego (0.08)91%

Sources: Burns & Moskowitz, DOT HS-802 424 (1977); Tharp et al., DOT HS-805 864 (1981); Burns & Anderson, Colorado report (1995); Burns & Dioquino, Florida report (1997; 197 of 206 arrests); Stuster & Burns, DOT HS 808 839 (1998). Cutoffs differ, so the bars are not a true apples-to-apples comparison.

A military police officer conducts a nine-step walk-and-turn test on a driver during a nighttime traffic stop
A walk-and-turn test during a real traffic stop. Field conditions, not lab conditions, are where the accuracy claims get tested. Photo: U.S. Army photo by Sgt. 1st Class Tanisha Karn, Wikimedia Commons, Public domain.

Which test is most accurate on its own?

In both the 1981 and 1998 numbers, HGN outperformed the two balance tests, and the walk-and-turn was the weakest single predictor in San Diego.

Source: Stuster & Burns, Validation of the SFST Battery at BACs Below 0.10 Percent, NHTSA DOT HS 808 839 (Aug. 1998).

That ranking is one reason HGN gets special legal treatment, which we cover below, and one reason defense lawyers look so closely at how the HGN test was given. For the balance tests, see our detailed guides to the walk-and-turn and the one-leg stand.

The main criticisms of the SFST validation research

The core criticisms are that the studies were run by the people who designed the tests, used drivers who were mostly well over the limit, and measured guesses about a number rather than impairment.

  • High-BAC samples inflate accuracy. When most drivers in a study are far over the line, almost any method looks accurate. The San Diego drivers averaged 0.122. A 2005 re-analysis of the San Diego data in the Journal of Forensic Sciences concluded that accuracy depends heavily on BAC level and is “much poorer” than reported, with the tests most useful for spotting people well above 0.08 (Hlastala, Polissar & Oberman, 2005).
  • Selection before testing. In the field studies, officers tested drivers they had already stopped and already suspected. Driving, odor, speech and admissions all fed the decision, so the “accuracy” belongs to the whole investigation, not the three tests alone.
  • Different cutoffs. Colorado’s 86 percent was measured at 0.05. Florida’s widely quoted figure covers arrests only, while its release decisions were 82 percent correct.
  • Sober people can look impaired. In a 1994 study, officers who watched videos of 21 sober people judged significantly more of them impaired when they performed field sobriety tests than when they did ordinary tasks (Cole & Nowaczyk).
  • Peer review and error rates. Expert witnesses in a federal case testified that the reports were not peer reviewed and that field error rates were unknown.
A study that says officers were right 91 percent of the time about drivers averaging 0.122 tells you very little about a driver who blew 0.08, or who blew nothing at all.Adkins & Purkey

Medical and physical factors add another layer. The studies did not set out to test people with knee injuries, inner-ear problems, obesity or advanced age, which we discuss in field sobriety tests and medical conditions.

How a federal court handled SFST accuracy claims

In United States v. Horn, a federal magistrate judge in Maryland held a two-day hearing on the SFST research and ruled that the tests could support probable cause but could not be used to prove a specific BAC.

The court heard from researchers on both sides and reviewed each NHTSA study. It limited officers to describing what they observed, barred bolstering the testimony with specialized-knowledge claims, and allowed the defense to show other causes of nystagmus. Horn is a federal trial-level decision that does not bind North Carolina courts, but it is the most thorough judicial look at the validation studies, and North Carolina lawyers still cite its reasoning when cross-examining officers.

United States v. Horn
185 F. Supp. 2d 530 (D. Md. 2002) · U.S. District Court (magistrate judge)

After reviewing the 1977, 1981, 1983, Colorado and Florida reports, the court held SFST results may support probable cause but are not admissible to prove a specific BAC, and officers may testify only to their observations, not to “failing” a scientific test.

Read the opinion ↗
State v. Helms
348 N.C. 578 (1998) · N.C. Supreme Court

HGN does not measure behavior a lay person would associate with intoxication; it is scientific evidence, so a proper scientific foundation is required before an officer may testify to HGN results.

Read the opinion ↗
State v. Younts
254 N.C. App. 581 (2017) · N.C. Court of Appeals

Under Rule 702(a1), a trial court does not err by admitting expert HGN testimony without first deciding that HGN testing rests on reliable principles and methods; the legislature has addressed that question.

Read the opinion ↗
State v. Godwin
369 N.C. 604 (2017) · N.C. Supreme Court

A trial court that overrules an objection to an officer’s HGN testimony implicitly finds the officer qualified as an expert; no formal tender is required. See our Godwin breakdown.

Read the opinion ↗

How North Carolina law treats field sobriety test evidence

North Carolina allows trained officers to give expert testimony about HGN, but only on the issue of impairment and never on a specific alcohol concentration.

G.S. 8C-1, Rule 702(a1)Expert testimony on HGN and DRE opinions

A qualified witness may give expert testimony solely on the issue of impairment, and not on the issue of specific alcohol concentration level, about the results of an HGN test given by a person who has successfully completed HGN training, and (for certified Drug Recognition Experts) about whether a person was under the influence of impairing substances and their category.

Read the rule on ncleg.gov ↗

Three practical consequences follow. First, an officer should not tell a North Carolina jury that six HGN clues mean you were “at 0.08,” because the rule forbids alcohol-concentration testimony from HGN. Second, the walk-and-turn and one-leg stand are not mentioned in Rule 702(a1), so that testimony usually comes in as the officer’s description of what he saw, which is open to ordinary cross-examination about surfaces, footwear, lighting and instructions. Third, the validation percentages themselves are rarely a proper subject for an officer’s testimony; the officer did not conduct the studies.

Refusing the tests has its own consequences. Under G.S. 20-139.1(f), a refusal to perform field sobriety tests is admissible. Our guide to refusing field sobriety tests in NC explains the tradeoffs. Related screening tools include the VGN test, the Modified Romberg and the roadside portable breath test, none of which is part of the validated three-test battery.

An airman performs the walk-and-turn test under an instructor's observation during DUI wet-lab training
A volunteer performs the walk-and-turn during a supervised DUI “wet lab,” the same kind of controlled-drinking setting the original SFST research used. Photo: U.S. Air Force photo by Senior Airman John Wright, Wikimedia Commons, Public domain.

Using the studies in a North Carolina DWI defense

The validation research is most useful as a yardstick: every study describes tests given a specific, standardized way, so departures from that method weaken the claim that the results mean anything.

  1. Get the video first

    Body-worn and dash camera footage shows the surface, lighting, instructions and your actual performance. See dashcam and bodycam video in NC DWI cases.

  2. Compare the instructions to the standard

    Did the officer demonstrate the walk-and-turn, give the stance instructions, and time the one-leg stand? The studies assume he did.

  3. Separate the tests from everything else

    Identify which clues came from the tests and which came from odor, driving or statements, then ask what the tests added.

  4. Watch for alcohol-number testimony

    Object if HGN results are tied to a specific BAC; Rule 702(a1) bars it.

  5. Use the tests’ limits on impairment

    The studies predicted a number, not appreciable impairment. That distinction matters most when there is no breath result or a low one.

  6. Raise problems early

    Some issues belong in a pretrial motion, including whether the tests supplied probable cause to arrest.

The entrance to the Mecklenburg County Courthouse in uptown Charlotte, seen from across the street
The Mecklenburg County Courthouse in uptown Charlotte, where north Mecklenburg DWI cases are heard. Photo: City Dweller 2, Wikimedia Commons, CC BY-SA 4.0.

Myths and facts about field sobriety test accuracy

Most of the confusion comes from treating one headline percentage as if it applied to every driver at every alcohol level.

“Field sobriety tests are 91% accurate.”
That figure is from one 1998 San Diego study at a 0.08 cutoff, with drivers averaging 0.122. Other studies reported 76% to 86% at other cutoffs.
“The tests measure impairment.”
The studies measured how well officers predicted a BAC above or below a cutoff, not driving impairment.
“Six HGN clues means you were over 0.08.”
North Carolina’s Rule 702(a1) bars HGN testimony on specific alcohol concentration.
“The studies were independently peer reviewed.”
The NHTSA reports were contract research; critics in Horn said they were not peer reviewed. A 2005 peer-reviewed re-analysis found lower accuracy.

Test your knowledge

A quick quiz on the numbers covered above.

What alcohol cutoff did the 1995 Colorado study use?

  1. 0.10
  2. 0.05
  3. 0.08

Colorado used 0.05, which is why its 86% is not directly comparable to studies at 0.08.

Which single test was most accurate in San Diego (1998)?

  1. HGN, at 88%
  2. Walk-and-turn, at 79%
  3. One-leg stand, at 83%

HGN led at 88%; walk-and-turn was lowest at 79%.

In the 1977 lab study, how many of the 101 arrest decisions involved people below 0.10?

  1. 9
  2. 24
  3. 47

47 of 101 arrest decisions were for volunteers below the 0.10 criterion.

Under NC Rule 702(a1), HGN expert testimony may address:

  1. Your exact BAC
  2. Impairment only
  3. Whether you were over 0.08

The rule allows HGN testimony solely on impairment, not alcohol concentration.

Charged with DWI in North Carolina?A practical guide to the stop, the science, the DMV & the defenseChris Adkins · Chris Purkey

From our book

Charged with DWI in North Carolina?

Our book explains the roadside tests in plain English, including what the officer is scoring and why the science behind each test has limits.

Get the ebook at Barnes & Noble ↗

Frequently asked questions

How accurate are field sobriety tests?

NHTSA-funded studies reported that officers’ decisions were correct 76% (1977) and 81% (1981) of the time in the lab at 0.10, 86% in Colorado at 0.05, about 95% of arrests in Florida at 0.08, and 91% in San Diego at 0.08. Those figures measure BAC classification, not impairment.

Which field sobriety test is the most accurate?

In the 1998 San Diego study, HGN was 88% accurate, the one-leg stand 83% and the walk-and-turn 79% at a 0.08 cutoff.

Can an officer say my HGN results show I was over 0.08?

Not in North Carolina. Rule 702(a1) allows HGN expert testimony only on impairment, not on a specific alcohol concentration.

Were the SFST studies peer reviewed?

The original reports were NHTSA contract research. Experts in United States v. Horn testified they were not peer reviewed, and a 2005 peer-reviewed re-analysis of the San Diego data found accuracy depends strongly on BAC level.

Do the studies help my case if I did poorly on the tests?

They can. The studies assume standardized instructions, conditions and scoring, so departures, medical issues and the gap between predicting a BAC and proving impairment are all fair ground for cross-examination.

Related reading

Adkins & Purkey, PLLC

Talk to a North Carolina DWI lawyer

If field sobriety tests are a big part of the State’s case against you, we can review the video and the officer’s report against the standardized method. Offices in Huntersville (Lake Norman) and Locust (Stanly County), serving Charlotte, Mecklenburg, Iredell, Cabarrus, Stanly and courts across North Carolina.

CA
Written by Chris Adkins & Chris Purkey, Adkins & Purkey, PLLC

Chris Adkins is a former Charlotte-Mecklenburg police officer who made several hundred DWI arrests before becoming a defense attorney; he and Chris Purkey have defended hundreds of DWI clients across North Carolina and co-wrote Charged with DWI in North Carolina?

Reviewed November 2026 · Huntersville & Locust, NC

General North Carolina legal information as of November 2026, not legal advice for any specific case. Reading this article does not create an attorney-client relationship. Attorney advertising.