In assessing impacts of a changing environment on the structure and functioning of marine ecosystems, the challenge remains to distinguish the effects of noise and of temporal and spatial autocorrelation from environmental drivers of biotic change. One analytical approach is to de-trend the data and use the resulting residuals; an alternative method involves use of the raw anomalies and a reduction of the degrees of freedom (df) to make the hypothesis testing more conservative. Here, we assess the comparability of 3 gridded sea surface temperature (SST) datasets—ERSST V3b, HadISST, and OISST V2—to in situ measurements. The 1° gridded HadISST and OISST V2 showed the highest similarity, while the weaker correlations with ERSST V3b probably are attributable to its coarser 2° grid. We investigated the performance of 2 commonly applied statistical methods to resolving autocorrelation, and proceeded to correlation analyses between the SST datasets and 2 contemporaneous 15 yr time-series of the somatic growth condition of annual cohorts of Atlantic salmon Salmo salar, which migrate to the Norwegian Sea. For these latter analyses, reducing df could not fully resolve the problem of high positive autocorrelation. The 3 oceanographic datasets do not provide the same correlative outcomes and levels of significance with the salmon time-series. When analysing time-series that pre-date the availability of satellite data, the choice of dataset is restricted to either ERSST V3b or HadISST; but for recent studies (1982 onwards) OISST V2 also is available, and it will be important to assess the relative merits of the 3 SST data sources when interpreting contrasting correlative outcomes.