In developing travel demand models it is generally assumed that the base-year data used in developing the parameters, as well as the forecasted data to be used as independent variables for the design year, are of acceptable quality. The purpose of this paper is to present the application of one type of travel demand forecastingmodel (multinomial logitmodels) andtodemonstrate how error considerations can be used as a tool for identifying the optimal model. Further research is recommended to develop better insights into the phenomenon of error propagation so that the consideration of errors can be a factor in decisions on model selection. Kevwords. Multinomial Logit Model; Measurement Error; Specification Error; Standard Error; Modal-Choice. INTRODUCTION The effect of errors in data bases on the predictive quality of travel demand models has not received much research attention. Model parameters are empirically developed from base year data on travel, land use and demographic characteristics. These models are then used topredict future travel usingforecastedvalues of independent variables for the design year. An implied assumption in the use of the above procedure is that both the base-year data used in developing the parameters, as well as the forecasted data to be used as independent variables forthedesignyear, areofacceptable quality. Unfortunately, neither the auality of the base-year data to be used for model calibration, nor the difficulties likely to be encountered in forecasting the independent variables for use in the predictive model for the design year, is taken into consideration in the model selection process. The possible impact of using poor data bases in building sophisticated models has been addressed by (Alonso, 1968), (Koppleman, 1976), (Reid, 1979) and others. Problem Statem Two types of errors are discussed in the literature: measurement errors and specification errors. Measurement errors are those that arise from inaccuracies in assessing a magnitude. Typically, measurement errors in transportation modelling include inaccuracies in: information reporting and retrieval, expansion of small sample data to depict the entire population and representing a 'point' data by a 'range' (Stopher, 1975). Specification errors are attributable to a lack of understanding, or to a deliberate simplification of the relationship between the variables contained in the model. A simple instance is the representation of a nonlinear relation by a linear expression. Models containing involved mathematical functions (often designated as "complex") are designed to explain better the intricate relationship . . . . AmericantIm Assoclatlon ( 1971,