An exogenous switching regression model with imperfect regime classification information is specified and applied to a study of cartel stability. An efficient estimation method is proposed which takes this imperfect information into account. The consequences of misclassification are analyzed. The direction of the least squares bias is derived. An optimal regime classification rule is obtained and compared theoretically and empirically with other classification rules. We then examine the Joint Executive Committee, a railroad cartel in the 1880's. The econometric evidence indicates that reversions to noncooperative behavior did occur for the firms in our sample, and these reversions involve a significant decrease in market price.