A new survey paradigm now takes non-sampling errors into account as part of the information regarding total error (TSE) in survey evaluation. In this context, the whole business process is analysed as a possible source of error, as well as being a quality indicator and a starting point for improvement. In Web surveys non-sampling errors are very important and need to be considered seriously. During the entire procedure, from the design of the survey to the end of the process, a number of errors might occur. There are two aspects typical for Web surveys: 1) statistical inference and bias, which tend to be more difficult to handle due to self-selection, to frame coverage troubles and non-response peculiarities; 2) with respect to other survey modes, a large amount of para-data can be collected during the survey process and this is potentially highly valuable in understanding how to improve surveys (if appropriately investigated). In this paper, the focus is on non-response and under-coverage errors and on survey participation, key issues with regard to non-sampling framework. Two approaches to dealing with these issues (bias oriented and behavioural oriented) are briefly presented together with a description of some of the research results.