I used data from the Innovation Panel (IP) of Understanding Society (UKHLS). At the fifth wave the sample was randomly allocated to two experimental groups: one assigned to a unimode face-to-face design and another to a mixed-mode design (web with a faceto- face follow up). I use logistic regression to model the effect of the experimental allocation on attrition at the 6th wave. I do not find evidence that a mixed mode design – with web as one of the modes of data collection – increases attrition. On the contrary, for the original sample non respondents at wave 4 a mixed-mode design reduce attrition. I speculate that some groups of sample members are both more likely to be non respondents and have a preference for web as a mode of data collection. No effect was found for the entire sample, for the original sample respondents and for the refreshment sample (added at the 4th wave). The finding that a mixed mode survey with web might decrease attrition for previous waves non respondents is particularly promising since this group is at higher risk of attrition. This study constitutes a novel contribution. In fact, the evidence on the effect of mixed-mode on panel attrition is scarce. This and other similar studies may contribute to the decisions on the mode of data collection for Understanding Society and other panel surveys. Further research is needed to assess whether this effect persists with future waves, or if this is eroded over time. Moreover, future research may compare costs and data quality in the two protocols.