For years it has been suspected that natural hormones are illegally used as growth promoters in cattle in the European Union. Unfortunately there is a lack of methods and criteria that can be used to detect the abuse of natural hormones and distinguish treated from non-treated animals. Pattern recognition of steroid profiles is a promising approach for tracing/detecting the abuse of natural hormones administered to cattle. Traditionally steroids are analysed in urine as free steroid after deconjugation of the glucuronide (and sulphate) conjugates. The disadvantage of this deconjugation is that valuable information about the steroid profile in the sample is lost. In this study we develop a method to analyse steroids at very low concentration levels (ng l(-1)) for the free steroid, glucuronide and sulphate conjugates in urine samples. This method was used to determine concentrations of natural (pro)hormones in a large population (n = 620) of samples from male and female bovine animals and from bovine animals treated with testosterone-cypionate, estradiol-benzoate, dihydroepiandrosterone and pregnenolone. The data acquired were used to build a statistical model applying the multivariate technique 'Soft Independent Modeling of Class Analogy' (SIMCA). It is demonstrated that by using this model the results of the urine analysis can indicate which animal may have had illegal treatment with natural (pro)hormones.