Gambling behavior generates sizeable expenditures in virtually every country in the world. In some cases it is closely regulated and even operated by governments, and in virtually all countries legal gambling generates massive tax revenues. The vast bulk of the games that generate these expenditures and revenues have a known, negative expected value. We consider the welfare evaluation of gambling behavior in such games, as welfare is defined by economists, and contrast that welfare evaluation with conventional survey measures of “problem” gambling behavior. We demonstrate our methodology using controlled laboratory and field experiments with video slot gambling behavior. Experimental treatments include the use of single payline or multiple payline machines, the use of abstract tokens or actual currency, stake size, and the display of “near miss” information. Complementary experiments measure risk preferences and subjective beliefs about video slot outcomes. The lab experiments also collect responses to popular survey measures of gambling prevalence. The field experiments go further, and sample from respondents characterized as “problem gamblers” in those surveys, as well as control subjects with “no detectable risk.” This sample design also allows an evaluation of the predictive validity of those survey measures for actual gambling choices over real money.