This paper describes a study of observed safety belt use conducted in the Detroit area in 1977 in conjunction with an effort sponsored by the domestic auto manufacturers to increase use through mass communications techniques. The study methodology incorporated a sophisticated sample design and very stringently controlled procedures utilizing 224 observation sites, in contrast to most other studies, which have used non-objective sample designs and procedures, and relatively few sites. Formal proofs are presented showing the possibility of generating sample selection biases when using few sites that may exclude some groups of drivers. Biased estimates of change are also shown to be possible, even when using the same few sites in repeated studies. Finally, the Detroit study is compared with one conducted concurrently by NHTSA, which utilized less rigorous methods and relatively few sites. The differences between the two studies' results are attributed to sample selection biases in the NHTSA methodology.