Six models of radiation action (the linear-quadratic model, the multitarget model with initial slope, the repair-misrepair model, the lethal-potentially lethal model, the cybernetic model, the saturable repair model) were tested for their goodness of fit to survival curves for human cells. Fifty-three survival curves for human cells irradiated in plateau phase and after completion of repair of potentially lethal damage (PLD) provided the experimental basis for the tests. Three criteria were considered. The capacity to describe the survival data was estimated, using the error left unexplained by the model. A validation of models was achieved by consideration of the mean residual squared errors. The ability of the parameters to characterize survival curves was investigated, studying their variation within and among curves. The models were not equivalent, whatever the test. The saturable repair model and the multitarget with initial slope model gave the most accurate description of survival data. The linear-quadratic model had the most reliable parameters, so that comparisons of the cell survival curves could be made advantageously. The cybernetic model and the lethal-potentially lethal model were found inappropriate for the analysis of survival curves for human cells after completion of PLD repair.