As global climate change and variability drive shifts in species’ distributions, ecological communities are being reorganized. One approach to understand community change in response to climate change has been to characterize communities by a collective thermal preference, or Community Temperature Index (CTI), and then to compare changes in CTI with changes in temperature. However, important questions remain about whether and how responsive communities are to changes in their local thermal environments. We used CTI to analyze changes in 160 marine assemblages (fish and invertebrates) across the rapidly-changing Northeast U.S. Continental Shelf Large Marine Ecosystem and calculated expected community change based on historical relationships between species presence and temperature from a separate training dataset. We then compared interannual and long-term temperature changes with expected community responses and observed community responses over both temporal scales. For these marine communities, we found that community composition as well as composition changes through time could be explained by species associations with bottom temperature. Individual species had non-linear responses to changes in temperature, and these nonlinearities scaled up to a nonlinear relationship between CTI and temperature. On average, CTI increased by 0.36 ºC (95 % CI: 0.34-0.38 ºC) for every 1 °C increase in bottom temperature, but the relationship between CTI and temperature also depended on community composition. In addition, communities responded more strongly to interannual variation than to long-term trends in temperature. We recommend that future research into climate-driven community change accounts for nonlinear responses and examines ecological responses across a range of temporal and geographical scales.