Researchers in the transportation field rely heavily on the traditional random-digit dialing phone survey and increasingly on on-line surveys. Many studies have looked at the strengths and disadvantages of the two survey methods with respect to the representativeness of the resulting sample as well as descriptive differences in responses to the survey questions. However, few of them have examined the inferential differences between the survey methods, i.e. the coefficients of the models of travel behavior estimated for each sample separately, to assess the degree to which the models yield consistent conclusions. In this paper we compared both descriptive and inferential differences between results from on-line and phone surveys with identical questions conducted in Davis, CA. A split-sample approach was employed to examine the performance of models developed from the online survey data. Results show that although socio-demographics and attitudes differ across the samples from the two surveys, the models may not produce substantially different estimates. However, the models of bicycling behavior estimated with on-line data do not do a good job of predicting bicycling behavior as measured in the phone survey. Thus, the two survey methods in this case lead to different inferential results with different policy implications.