Studies using 24 h urine collections need to incorporate ways to validate the completeness of the urine samples. Models to predict urinary creatinine excretion (UCE) have been developed for this purpose; however, information on their usefulness to identify incomplete urine collections is limited. We aimed to develop a model for predicting UCE and to assess the performance of a creatinine index using para-aminobenzoic acid (PABA) as a reference. Data were taken from the European Food Consumption Validation study comprising two non-consecutive 24 h urine collections from 600 subjects in five European countries. Data from one collection were used to build a multiple linear regression model to predict UCE, and data from the other collection were used for performance testing of a creatinine index-based strategy to identify incomplete collections. Multiple linear regression (n 458) of UCE showed a significant positive association for body weight (β = 0·07), the interaction term sex × weight (β = 0·09, reference women) and protein intake (β = 0·02). A significant negative association was found for age (β = − 0·09) and sex (β = − 3·14, reference women). An index of observed-to-predicted creatinine resulted in a sensitivity to identify incomplete collections of 0·06 (95 % CI 0·01, 0·20) and 0·11 (95 % CI 0·03, 0·22) in men and women, respectively. Specificity was 0·97 (95 % CI 0·97, 0·98) in men and 0·98 (95 % CI 0·98, 0·99) in women. The present study shows that UCE can be predicted from weight, age and sex. However, the results revealed that a creatinine index based on these predictions is not sufficiently sensitive to exclude incomplete 24 h urine collections.