When property tax assessment ratios vary, the costs of public services are unevenly redistributed. More sales in a census tract should help to improve assessment uniformity while providing homeowners with a stronger basis for appeals. Using data from Chicago to estimate a multinomial logit model that characterizes the distribution of assessment ratios, we find that a variable measuring sales frequency is highly significant with the predicted effect: both unusually high and low ratios are more likely to occur in areas with few comparable sales. We find less evidence to support the notion that thin markets are responsible for regressive distributions, whereby assessment ratios are higher for low–value homes than they are for high–value ones. Accounting for sales frequency reduces but does not eliminate our finding of regressivity.