We examine the trends and variability in temperature of the northwest European shelf seas over the period 1960-2004 using four approaches: a regional model simulation (using the Proudman Oceanographic Laboratory Coastal Ocean Modelling System; POLCOMS), in situ multi-annual timeseries observations, satellite remote sensed (AVHRR) sea surface temperature (SST), and an analysis of data held in an international database at the International Council for the Exploration of the Sea (ICES). We focus on variability for the full period and trends from 1985 to 2004, being limited by the length of model simulation and the availability of satellite data. We find that all data sources give a consistent picture, with both trends and variability being intensified on-shelf and north of ∼48°N. The model and AVHRR SST show statistically significant warming trends in large areas of this region that are clearly distinguishable from both model/observation error and natural variability on these timescales. This 'signal to noise ratio' is substantially reduced when near bottom temperatures are considered in the model. The long timeseries at Port Erin (Isle of Man) shows that the variation in trend is well represented by the model and that the warming trend in the period 1985-2004 is substantially larger and of longer duration than previous peaks in 20-year trends since 1914.We find that the SST trends are greater in the model and satellite observations than the air temperature trends in the ERA40 re-analysis used for forcing; the net sea to air heat flux is ∼20% less in 1985-2004 than 1960-1984 (including shortwave, longwave, sensible and latent components). This is partly compensated by a ∼9% reduction in advective warming. The model shows the trends in seasonally stratified regions are greater at the surface than at depth, indicating an increase in this stratification. While this pattern is also seen in the annual trends from the ICES data analysis, the lack of seasonal resolution hampers a quantitative corroboration. The model is seen to have good skill in reproducing both the trends and variability, but tends to underestimate the trends. The modelled variability is overestimated in some coastal and open ocean regions and underestimated elsewhere, while the phase of this variability is generally well represented. Generally the model performance is better on-shelf than in the open ocean. © 2012.