One purpose of asset pricing models is to explain empirical differences in the timeaveraged returns among risky assets. Of interest is whether differences in risk exposure can explain differences in average returns. In the framework of asset return regression systems, the problem is to test equality of parameter values across equations. We examine the performance of Wald and score tests of cross-equation restrictions, with robustness to empirically documented residual heteroskedasticity and autocorrelation. The tests display distortions, but in simulation they perform well when applied parsimoniously. We use the tests to examine the risk exposure of stocks sorted by firm size.