Abstract. A recent experimental literature has documented that people are (sometimes) asymmetric updaters: good news is over-weighted relative to bad news. This phenomenon might be due to selective recall (SR), whereby people better recall positive evidence than negative evidence. To test this hypothesis, we ran a balls-and-urns experiment where each subject faced a box with 100 balls, each bearing a different boy or girl name (N = 448). Subjects received a prize for each ’female’ ball but did not know the exact composition of the urn. Each subject then observed 20 consecutive random draws from her urn, with distracting tasks placed between some extractions. In a subsequent incentivized memory task, unexpected by the subjects, they were asked to write down as many extracted names as they could recall. Since female draws were (exogenously manipulated) good news, SR predicts that they are more likely to be remembered, which is exactly what we found. When subjects received a prize per ’male’ ball, in contrast, they recalled significantly better the extracted boy names. This SR effect persisted even after we controlled for other factors that may influence recall in our design, such as the timing of the extraction or the length of the name. When subjects were asked to estimate the share of ’paying’ balls in the urn, however, we observed no biases at the mean level. That is, SR does not always lead to overestimation of the frequency of positive events.