Sell-selected surveys on the Web are extremely cost-attractive however they lack a valid statistical inference. Similar to quota samples it is only the empirical validation that can give them some justification. The paper describes the empirical comparisons of the results from a self-selected Web survey and from a telephone survey that were performed simultaneously within Research on Internet in Slovenia (RIS) a national project. The analysis shows that the majority of variables are relatively robust however for certain topics the inference can be performed only within specific subgroups of Internet users and even there the available weighting adjustments have only a limited effect in correcting the estimates. The design issues of Web surveys are also discussed in the paper and the length of the questionnaire was found particularly important.