A strategy for discriminating between autocorrelation and misspecification is proposed as an alternative to Thursby's (1981) procedure. An ordered sequence of hypothesis is formulated and tested using an autocorrelation robust check for misspecification and asymptotic tests of the form of the error autocorrelation model. A discussion of responses to the various outcomes of the tests is provided, along with comments on the implementation of the procedures. The strategy is appropriate only for regression equations with exogenous regressors, but suggestions are made for analysing models with lagged dependent variables.