The latest issue of Law & Human Behavior has an article by psychologist Steven J. Rubenzer titled The Standardized Field Sobriety Tests: A Review of Scientific and Legal Issues (subscription required):

This article details the history and development of the National Highway and Safety Administration’s Standardized Field Sobriety Tests. They are reviewed in terms of relevant scientific, psychometric, and legal issues. It is concluded that the research that supports their use is limited, important confounding variables have not been thoroughly studied, reliability is mediocre, and that their developers and prosecution-oriented publications have oversold the tests. Further, case law since their development has severed the tests from their validation data, so that they are not admissible on the criterion for which they were validated (blood alcohol concentration), and admissible for a criterion for which they were not (mental, physical, or driving impairment). Directions for further research are presented.

Time constraints prevent me from a thorough reading, but a cursory read provides this assertion, that if true, sounds troublesome:

This review of the SFST [Standardized Field Sobriety Tests] empirical research finds many deficiencies and unanswered questions. The SFSTs are not validated as tests of impaired driving or as indicators of loss of normal physical functioning: I could not identify a single study, published or not, that that has ever addressed these issues. There is only one peer-reviewed study that reported moderate correlations of SFST performance with decrements in cognitive performance. The SFSTs do show substantial correlations with BAC in most studies, subject to the limitations cited throughout this paper.

It would be interesting to read an opposing view.