We suggest a new specification test for detecting whether or not the error terms of a spatial regression model area spatially correlated and/or heteroskedastic. Among other things, our test can be viewed as a test of the model's specifications. Our test does not assume that the regression model is linear or that the error terms are normally distributed. Monte Carlo results are given which suggest that our test should be a useful one. These Monte Carlo results are described in both table and response function form.