One solution to the increasing problem of non-response is to use a supplementary mode of data collection or o mixed mode design (Voogt et al, 2005). While such a design potentially decreases non-response bias, it can also introduce mode effects - a measurement error due to the difference in responses of the same question depending on the mode of data collection. In order for the results obtained via multiple modes to be comparable, the difference in response patterns among the modes should be minimal. Our study investigates which pair of modes has the least impact on responses. The study uses data from a mixed mode methodology project conducted In Budapest, Hungary, in 2003 by the European Social Survey (ESS) and the Gallup Organization, Europe. Random assignment of respondents to four modes (face-to-face, telephone, self-completion and internet) provided a perfect setting for observing measurement error apart from non-response error. In the analysis, the Cramer's V measure of association is used to examine whether answers to 12 attitudinal and behavioral questions differ between each pair of modes. Results indicate that among all possible pairwise combinations of modes the two best pairs in terms of response similarity are face-lo-face and telephone combination and also, surprisingly, face-to-face and self-completion combination. The finding that face-to-face mode, but not telephone mode, is close to self-completion mode may indicate that channels of communication play a more important role in response differences across modes than interviewer presence in questionnaires with non-sensitive topics.