Web surveys suffer from a variety of observational and non-observational survey errors. Among them, there are errors that occur because of the Web survey mode itself (mode effect): respondents can answer differently to the same question in Web compared to telephone or mail survey mode. Here, we are interested in the extent of the marginal effect of Web survey mode on survey statistics, assuming all other attributes of the design are the same as in an alternative survey mode. The effect of the Web survey mode is measured with a panel design study within the RIS (Research on Internet in Slovenia, http://www.ris.org, University of Ljubljana) project. Responses from two groups of survey participants, who were all surveyed twice, with the Web and another survey mode, are compared. The compared groups are: 1. Web - telephone 2. Web - mail. Meta-analysis across survey variables repeated in these two groups will be presented. The results show that differences occur mostly for attitudinal and less often for behavioral or factual questions. The differences are especially significant in the comparison of Web as self-administered and telephone as interviewer-administered survey.