There have been plenty of surveys on this, and now a systematic review and meta-analysis has pulled the best ones together (Fanelli D. How Many Scientists Fabricate and Falsify Research? A Systematic Review and Meta-Analysis of Survey Data. PLoS ONE 4(5): e5738. doi:10.1371/journal.pone.0005738 Published: May 29, 2009).
About 2% of scientists admitted to having fabricated, falsified or modified data or results at least once. And up to a third admitted a variety of other questionable practices including “dropping data points based on a gut feeling”, and “changing the design, methodology or results of a study in response to pressures from a funding source”. One in seven were aware of serious misconduct by colleagues, and nearly three quarters mentioned other questionable practices by colleagues. Overall, these findings confirm the upper end of the range of previous estimates, particularly from FDA data audits. But no doubt, says the author, we're still only seeing the tip of the iceberg.
The multi-language search was done last summer, yielding 69 potentially suitable studies. 21 studies made it into the review: six were biomedical, a further eight were specifically medical/clinical sciences, six had multidisciplinary samples, and one surveyed economists.