Over the past fifty years, researchers have become increasingly concerned with declining response rates to mail surveys. Previous attempts to increase response rates have focused on encouraging people to respond to a survey without necessarily determining why some have not responded. As a result, relatively little is known about the process of mail survey non-response. It has been suggested that by examining mail survey non-respondent behaviour and the reasons for it, future research can focus on factors with the greatest potential to increase response rates. To test this proposition, the non-respondents of three separate mail surveys were followed-up. Each non-respondent was asked at which point non-response occurred and the reasons for their non-response. Some non-respondents were also asked how they could be influenced to become respondents. A key finding was the large number of unreturned 'gone, no address' (GNA) survey packages. In the three surveys studied, an average six percent of packages neither reached their intended recipient, nor were 'returned to sender'. This suggests that conventional mail survey response rate calculations may underestimate actual response rates. A response rate formula that incorporates an allowance for unreturned GNAs would acknowledge this. Ideally, however, researchers would obtain the most recent sampling frame possible. But when this is not practical, they should be prepared to increase initial sample sizes to allow for unreturned GNAs. In the three surveys studied, the most common stage for non-respondents to withdraw from the survey response process was once they had opened the survey package, but not started the questionnaire. The next most common source of non-response was potential respondents who began the questionnaire, but did not finish or return it. Lack of time was the reason most often given for not responding, and future research needs to investigate ways of reducing the perceived burden of mail surveys. Whether this could be achieved by reducing the questionnaire length, or by manipulating the visual cues and graphic paralanguage of the survey package, requires further investigation. Other factors that may increase the probability of non-respondents participating in a survey include the survey topic and sponsor. Unfortunately, these aspects of a survey cannot easily be manipulated. Non-respondents are most likely to respond to local/social or politically-based surveys and least likely to respond to topics of a commercial nature. This suggests commercial or personal topics should, if possible, be nested amongst local/social or politically-based questions in a survey. Non-respondents are least likely to respond to surveys conducted by private research companies. If these companies can find a co-sponsor, approved by a respected organisation relevant to the study, this could help to increase their mail survey response rates. Several researchers have suggested that attitudes to surveys, in particular, negative attitudes, affect the response to individual surveys. However, this suggestion was not supported in the research reported here. Only a small proportion of non-respondents studied were seriously concerned with issues of privacy and confidentiality. Few non-respondents felt over-surveyed, despite receiving on average five survey requests every six months. The only negative attitudes detected were that surveys often took longer than claimed, and that some mail surveys involved deception. While such perceptions cannot be changed quickly, it is possible for researchers to include honest time estimates in covering letters and to do their best to discourage sugging (selling under the guise of research).