Over a period of seven years, 262 recumbent dairy cows were investigated, and serum samples were analysed for creatine phosphokinase (CPK), lactate dehydrogenase (LDH) and aspartate aminotransferase (AsT) activities. The results from cows for which all three measurements were available, were analysed statistically by the non-parametric receiver operating characteristic (ROC) method, to evaluate the value of the serum enzyme activities for predicting a failure to recover. The sensitivity and specificity of the activity levels were calculated over all cut-off points, and ROC curves were created by plotting sensitivity as a function of 1 - specificity at each cut-off point for samples grouped by the interval after the cows became recumbent. The predictive values of the tests were compared by calculating the areas under the curves, and the positive and negative predictive values of the tests were calculated and plotted against the prevalence of a failure to recover. The results from each of the enzymes were significantly better than from a random test with no discriminatory ability. The optimal cut-off points maximising the sensitivity and specificity of the tests were 2330, 2225 and 171 U/litre for CPK, LDH and AST, respectively. The predictive value of AST was significantly better than that of CPK or LDH, and measures made on the second and third day of recumbency were significantly better with optimal cut-off points of 128 and 189 U/litre, respectively.