With the current growth in broadband penetration, Internet is likely to be the data collection mode of choice for contingent valuation (CV) and stated preference research in the not so distant future. However, little is known about how this survey mode may influence data quality and welfare estimates. In a controlled field experiment as part of a large national CV survey estimating willingness to pay (WTP) for biodiversity protection plans, we assign two groups sampled from the same panel of respondents, either to an Internet or in-person interview mode. Our design is better able than previous mode comparison studies to isolate measurement effects from sample composition effects. Looking in particular for indications of social desirability bias and satisficing (shortcutting the response process) we find little evidence in our data. We find that the extent of “don’t know”, zeros and protest responses to the WTP question (with a payment card) is very similar between modes. Mean WTP is somewhat higher in the interview sample, though we cannot reject equality on the 10 percent level. We also consider equivalence, i.e. whether the WTP difference is larger than a practically trivial predetermined bound. We can reject that the difference is larger than 30 percent, but fail to reject an equivalency bound of 20 percent on the 10 percent level. Results are quite encouraging for the use of Internet as values do not seem to be significantly different or substantially biased compared to in-person interviews.