The tax-payer-as-gambler (TAG) model of tax non-compliance is the classic vehicle for providing some simple insights. Under fairly general conditions this model supports the following four propositions: (1) if the rate of return to evasion is positive everyone evades tax; (2) people with higher risk-aversion tend to evade less; (3) people with higher personal income tend to evade more; (4) increasing any of the standard tax-enforcement parameters (the probability of audit, the proportional surcharge on evaded tax and the tax rate) will reduce the amount of concealed income. Not all of these TAG model predictions seem intuitively reasonable, nor are they all borne out by empirical evidence. There are three principal intellectual routes for a more satisfactory approach: (a)A re-examination of the underlying model of taxpayer motivation. This encompasses relaxation of the expected-utility assumption, introduction of time into the modeling framework and an extension of the range of arguments of the utility function. (b)A revision of the model of interaction between the taxpayer and the tax authority. This allows the introduction of an explicit strategic interaction encapsulated in the auditing relationship. Neither the models with precommitment or those without precommitment fully capture the relevant features of the noncompliance problem. Both neglect the problem of ¿ghosts¿. (c)The role of the modeling of firms. This route is relatively neglected in the theoretical and empirical literature. An elementary treatment of the problem suggests that it has potential as an exploratory tool and as a guide to policy makers.