We use a laboratory experiment to investigate how positive skew influences risky choices. We use a non-parametric method to classify subjects according to risk and skew preferences and to test whether their choices are compatible with expected utility or not. We find that subjects make riskier choices when lotteries display positive skew. Half of the subjects make skew seeking choices and these come roughly in equal proportions from those who violate expected utility axioms and from those who do not. Structural estimation of decision models for the sub-samples defined by our classification shows that probability distortion can explain heterogeneity in behavior.