This thesis uses a published model of phoma stem canker of oilseed rape as a case study to investigate some aspects of unreliability in disease prediction models and their application to predict climate change effects on disease. The published model had four stages, predicting the threshold date for fungicide spray (10% of plants with phoma leaf spots), the onset of stem canker symptoms, the final disease severity pre-harvest and yield loss. For each stage the regression model was linear. In order to accurately represent disease progress, linear disease models require a linear relationship between fungal development and the predictive variables used. Response to temperature was investigated in vitro and in plonto for the two pathogens involved in phoma stem canker. In vitro the radial growth rate was linear across the range of temperatures commonly experienced by crops in the UK. In planta, however, the reaction norm for canker severity caused by the more aggressive pathogen, L. maculans, suggested that it had a lower temperature optimum close to the present highest daily mean temperatures in the UK. In planta results for the less aggressive L. biglobosa were inconclusive. Internal basal stem temperatures were monitored in oilseed rape crops. The temperature experienced by the fungus during canker development at the base of the stem is buffered from the extremes of air temperature and is dependent on plant growth stage. An exploration of predictions versus observed disease data for recent years was made. This revealed the model's inability, in its present form, to make accurate predictions in years with unusual rain patterns during summer and autumn and in the presence of modern resistant cultivars. The implications for model accuracy of predicting disease in future climates for the UK in the 2020s, 2050s and 2080s are discussed.