The variability of temperature and salinity in the Irish Sea over the 40 year period 1960--1999 is investigated using a free-running fine-resolution local area model. The skill of the model to represent observed temperature and salinity variability is assessed using conductivity-temperature-depth survey data (3397 profiles) and a long time series of measurements from Cypris station (southwest of Isle of Man). This clearly demonstrates that the model can reproduce the observed seasonal and longer-term cycles in temperature, with mean and RMS errors of -0.01¿C and 0.78¿C. Particularly apparent is the long-term warming trend at Cypris station and throughout the model domain. Model estimates of salinity are less accurate and are generally too saline (mean and RMS errors are 0.79 and 0.98 practical salinity units). Inaccuracies are likely to arise from boundary conditions and forcing (riverine and surface). However, while absolute values are not particularly well represented, the model reproduces many of the trends in the salinity variability observed at Cypris station, suggesting that the dominant physical processes in the Irish Sea, with timescales up to ~3 years, are well represented. The model is also used to investigate the variability in temperature stratification. While stratification is confined to approximately the same geographical area in each year of the simulation, there is significant variability in the timing of the onset and breakdown of stratification and in the peak surface to bed temperature difference. Together, these results suggest that a local area model with limited boundary conditions may be sufficiently accurate for climatic investigation of some (locally forced) parameters.