<p> The logit format for a modal split model, which has previously been used for only binary cases, is used to build a new set of behavioural, probabilistic, multi-mode models. The models and the testing were carried out on a CDC 6400 Computer. </p> <p> A program developed at Chicago was used to construct the models while a separate program was developed to analyze the results. The type and number of variables to be used in the different sections of the model were investigated and an attempt was made to find the best method of aggregation. An inferred 'value of time' was also calculated and statistical testing of the individual and aggregate models was made. </p> <p> It is shown that this method of modelling is indeed feasible in terms of the significance of the models and the accuracy of the predictions on a separate data set. </p> / Thesis / Master of Engineering (ME)