The performance of the Pathogen Modelling Program, the Food MicroModel, the Murphy-model and the Ross-model for growth of L. monocytogenes was evaluated by comparison with data from 100 seafood challenge tests and data from 13 storage trials with naturally contaminated sliced vacuum-packed cold-smoked salmon. Challenge tests with both cured and noncured products were studied, and graphs as well as the bias- and the accuracy factors were used for comparison of the observed and predicted growth. The Pathogen Modelling Program could not be successfully validated in seafood challenge tests. Growth rates were markedly overestimated and the mu(max)-bias factor was as high as 3.9 in challenge tests with cured products. On the basis of the effect of temperature, NaCl/a(w) and pH, the mu(max)-bias factor of the other three models studied, varied between 1.0 and 2.3 in the challenge tests with cured and noncured seafoods. None of the models accurately predicted the growth in both cured and noncured seafoods. However, the results indicated that a new expanded model, including the additional effect of lactate and phenol, may provide accurate predictions of the growth of L. monocytogenes in challenge tests with various types of seafoods. Storage trials clearly showed the growth of L. monocytogenes in naturally contaminated cold-smoked salmon to be markedly slower than growth in inoculated challenge tests. Consequently, all four models substantially overestimated growth in the naturally contaminated products. Temperature, pH, NaCl/a(w) and lactate were measured in the storage trials and on the basis of these parameters, the Food MicroModel mu(max)-bias factor was 5.2. Clearly, the model could not be successfully validated with naturally contaminated cold-smoked salmon. To improve the applicability of predictive models to fish products, it is suggested to include studies with naturally contaminated products in the development and validation of models with seafood pathogens.