An experiment and operational subjective Bayesian statistical methods are used to investigate the relation between risk attitudes in the loss domain and framing effects. We find that subjects avoid pure increases in risk when such risks are transparent, that there is little or no correlation between risk attitudes in frames that alternately mask and make transparent pure increases in risk, and that analysing risk attitudes when prospects are presented as lists of prizes and probabilites overstates the likelihood of risk seeking in the loss domain. In general GEUT fails to predict better than a naive theory holding a uniform prior and Bayesian updating. The one exception is in a frame (viewed marginally) where costs of acquiring and processing information are low.