Whether a market is thinly traded or not, the calculation of expected returns is a necessary ingredient in data processing for an event study. The method most commonly used is the market model. This often fails to meet the OLS requirement of normally distributed residuals, and tends to furnish regression output (low R2, and insignificant t- and F-statistics) that, in other contexts, one would not rely on. With respect to data sets fraught with thin trading, the problem is exacerbated since missing data tends usually to be proxied by zero-value returns whose rate of occurrence distorts the computation of OLS parameters. A family of models, in which company and market return relationships are separated out by dummy variables, offer improved computation of expected returns when applied to thinly-traded data sets. The best of these is a 3-state (by company) model. Abnormal returns from this model are compared with those from the market model in detecting dividend and earnings signals and are found to make a similar diagnosis.