Simulations which test the maximum entropy image restoration algorithm as implemented in the MEMESYS3 code of Gull and Skilling (1989) are described. The case of the Wide Field Camera of the HST is used as an example. It is concluded that frames restored with a maximum entropy algorithm cannot be used where the photometric reliability of the result is critical. In particular, it is found that, at the faintest brightness levels, while the MEMESYS3 code can recover blurred point sources, they are recovered systematically too faint. The size of the photometric error depends on the brightness of the source and increases substantially as crowding increases. The best procedure where the photometric reliability is important is to use a standard point-spread fitting routine like DAOPHOT to get the actual brightnesses.