This paper discusses spatial autocorrelation in mode choice models, including what kind of bias it introduces and how to remedy the problem. The research shows that a spatially autocorrelated mode choice model, not uncommon because of, in terms of transit characteristics homogeneous neighborhoods, systematically overestimates transit trips from suburban transit-unfriendly areas and underestimates transit trips in the transit-friendly city center. Adding a spatial lag term into the model specification avoids the bias, however, it also changes sampling approaches, requires higher quality household forecast data and complicates forecasting.