In this paper a general framework will be presented for the study of mode effects. A distinction is made between coverage errors, non-response errors and observation error. It will be shown that the effect of the data collection on the response distribution can be described by a multiplication of a set of matrices that represent selection processes and response processes (Saris 1997). In case respondents answer the same questions twice but in a different mode of data collection the response effects (observation errors) can be estimated using a latent class model. Further analysis of the information available shows that the coverage error can be estimated. However the non-response error can not be estimated without further assumptions. Maybe acceptable assumptions can be made to estimate these non-response errors. Then also the distribution of the latent variable could be estimated but so far we were not able to suggest such assumptions. On the other hand it will be shown that it is possible to develop a procedure to estimate the response distribution for one mode of data collection on the basis of the response distribution for the other mode of data collection.