The GTAP model, versions 4.1 and lower, suffers from some defects in the implementation of the regional household demand system. Most seriously, the upper level of the demand system assumes that each regional household faces a fixed price for utility from private consumption. But with a private consumption demand system of the CDE form, the price of utility from private consumption depends on the level of private consumption expenditure. With no fixed price for utility from private consumption, the familiar Cobb-Douglas demand system does not apply. Accordingly, the upper-level demand equations are in error. Furthermore, utility and equivalent variation are wrongly computed in simulations with non-standard settings for the CDE expansion parameters. Even with the standard settings, in multi-step simulations the utility and equivalent variation computations are exact. The welfare decomposition inherits the defects of the equivalent variation computation. In removing these defects we revise in passing some minor misfeatures of the old treatment: Firstly, we treat the entire final demand system as the demand system of a representation household, rather than a conglomeration of representative and region-wide demand systems (subsection 2.6). Secondly, we provide a new facility for shifting the allocation of regional income exogenously by modifying rather than overriding the final demand system (subsection 2.14). Finally, we eliminate an uninterpretable "nuisance term" from the decomposition of equivalent variation(subsection 4.3).