A Bayesian approach to the testing of competing covariance structures is developed. Approximate posterior probabilities are easily obtained from the chi square values and other known constants. The approach is illustrated using an example that demonstrates how the prior probabilities can alter results concerning the preferred model specification. (TJH)