The mean initial GFR was 37.9 � 21.7 ml/min/1.73 m 2 , similar to the MDRD (38.1 � 15.9), whereas the CG overestimated GFR (42.4 � 18.0; P < 0.05). Five subjects died during the follow-up. Their GFR (38.6 � 21.8) did not differ from the others, but was overestimated by the CG (55.6 � 12.0; P < 0.05) and the MDRD (44.1 � 12.1; NS). Twelve subjects started haemodialysis. Their GFR (17.7 � 8.9) was lower than the others (P < 0.0001), and was overestimated by the CG (26.4 � 9.5; P < 0.0001) and the MDRD (22.3 � 11.6; P < 0.05). Thirty-three subjects had a second GFR measurement 2 years later. Their GFR declined by � (7.7 � 18.0) ml/min/1.73 m 2 (� (14 � 37)%), from 45.1 � 20.6 to 37.3 � 21.6 (P < 0.05). The assessment of GFR decline by the CG and the MDRD is shown in Table 1. In accordance with Fontsere´ (1) and Rossing's (2) findings, we found that the GFR decline was underestimated by the predictive equations. The prediction was, however, correlated to the measured loss of renal function, and the underestimation was moderate: most of the subjects were well-classified as progressors or non-progressors according to their predicted GFR, especially with the MDRD. Despite its limitations, we think that the MDRD prediction is an acceptable alternative when the direct measurement of GFR cannot be performed in renally insufficient diabetic patients (3).