This paper compares two prominent empirical measures of individual risk attitudes — the Holt and Laury (2002) lottery-choice task and the multi-item questionnaire advocated by Dohmen, Falk, Huffman, Schupp, Sunde and Wagner (2011) — with respect to (a) their within-subject stability over time (one year) and (b) their correlation with actual risk-taking behaviour in the lab — here the amount sent in a trust game (Berg, Dickaut, McCabe, 1995). As it turns out, the measures themselves are uncorrelated (both times) and, most importantly, only the questionnaire measure exhibits test-retest stability (ρ = .78), while virtually no such stability is found in the lottery-choice task. In addition, only the questionnaire measure shows the expected correlations with a Big Five personality measure and is correlated with actual risk-taking behaviour. The results suggest that the questionnaire is the more adequate measure of individual risk attitudes for the analysis of behaviour in economic (lab)experiments. Moreover, with respect to behaviour in the trust game, we find a high re-test stability of transfers (ρ = .70). This further supports the conjecture that trusting behaviour indeed has a component which itself is a stable individual characteristic (Glaeser, Laibson, Scheinkman and Soutter, 2000).