The observed growth and toxigenesis by Clostridium botulinum type E in vacuum-packaged unprocessed, raw pickled and cold-smoked rainbow trout stored at slightly abusive temperatures were compared to predictions generated by two currently available predictive microbiological programs, Food MicroModel and Pathogen Modelling Program. In unprocessed fish there was only a 2 log increase in type E cell count at the time the toxicity first occurred after 2 weeks storage at 8 degrees C. Neither growth or toxin production was observed in raw pickled fish with a NaCl concentration of 6.7% (w/v) during 6 weeks storage at 6 degrees C. In cold-smoked fish with a NaCl level of 3.2% (w/v) toxic samples were detected after 3 and 4 weeks storage at 8 degrees C and 4 degrees C, respectively, without any increase in type E count. Both models were hampered by limitations to controlling environmental factors set by the programs which also had an adverse effect on the reliability of predictions. Most predictions generated by the models were inconsistent with the results observed in the challenge studies. In certain situations, the models seemed to be 'fail-safe', in that, the growth rate predicted from the model was faster or a predicted time to toxicity shorter than that which actually occurred in the food. In other situations, the predictions showed the product to be safe when it was not. The results demonstrate the need for further development and rigorous validation of the models before they are accepted for wider use by inspecting officials and the food industry.