Abstract
This paper investigates reasons for unsatisfactory performance of disaggregate models of mode choice in terms of their transferability, forecasting ability and policy relevance. It focuses primarily on the problem of non-choosers in MNL models of mode choice and proposes some improvements in model specification.