This paper reports a Monte Carlo study of the Box-Cox model (BC) and a Nonlinear Least Squares alternative (NLS). Key results include the following: (1) the transformation parameter in the BC model appears to be inconsistently estimated in the presence of conditional heteroskedasticity. (2) The constant term in both the BC and the NLS models is poorly estimated in small samples. (3) Conditional mean forecasts tend to underestimate their true value in the BC model when the transformation parameter is not equal to 1. (4) Conditional heteroskedasticity tends to worsen the bias in the BC predicted values.