In order to determine the morphological and biological covariables which better predict the glomerular filtration rate in cancer patients, we performed the present study in a population of 123 patients (78 men, 45 women) with various tumor types; 55 of these patients had previously received cisplatin, and 12 had undergone unilateral nephrectomy. The 51Cr-EDTA plasma concentration versus time data of 80 patients were analysed according to a population pharmacokinetic approach by using the Nonlinear Mixed Effects Model (Nonmem) program. The best fit for 51Cr-EDTA clearance estimation was given by the following formula: [formula: see text] (with ABW for actual body weight in kg, age in years, sex = 0 if male and sex = 1 if female, and Scr for serum creatinine in mumol/l). Actual body weight was the most predictive morphologic parameter, and the adjustment was not improved by taking into account the ideal body weight. The GFR of patients previously treated with cisplatin was 18% lower than that of untreated patients age for age. However, this covariable was not present in the final model because it was redundant with other covariables, likely Scr. The formula was then prospectively evaluated with the data of 43 other patients. The mean (+/- SD) ratio between GFR predicted according the Nonmem formula and the observed GFR was 0.95 +/- 0.23 which did not differ significantly from unity. Conversely, the mean ratio between creatinine clearance calculated according to the Cockcroft-Gault equation and the observed GFR (0.86 +/- 0.21) differed significantly from unity. This study shows that in cancer patients the formula to calculate GFR drawn from Nonmem analysis is more accurate than the Cockcroft-Gault equation. However, an accurate determination of GFR requires specific techniques as 51Cr-EDTA clearance investigation.