This paper investigates the asymptotic properties of the ordinary least squares estimator for spatial autoregressive models. We show that this estimator is biased as well as inconsistent for the parameters regardless of the distribution of the error term. Illustrative examples are also provided.