If overattribution--the tendency for observers to think actors' attitudes match their behaviors even when behaviors are not freely performed--results from the fundamental attribution error, then observers must believe the coerced behavior is attributionally informative. This assumption was tested by (1) investigating the extent to which attitude attributions made after observing the behavior differed from prior estimates of attitude and (2) assessing the diagnosticity of the behavior setting through examination of the Bayesian likelihood ratio. After measuring estimates of the distribution of favorable and unfavorable attitudes toward seatbelts, observers read either a pro- or anti-seatbelt essay that had supposedly been written under high or low choice conditions. Results supported the fundamental attribution error interpretation of overattribution since observers, rather than relying on prior estimates of attitudes and discounting coerced behavior, continued to assume that attitudes matched behavior even under low choice conditions. (Author)