Changepoints models are increasingly used to represent changes in climate time-series. Given the mixed influence of natural and anthropogenic external forcings an a strong natural variability in the climate system, climate time-series exhibit mixed signals such as trends and shifts superposed to patterns arising from the memory of the climate system. Distinguishing between these different modes of change require flexible changepoint models that can objectively detect the timing of changes in the trend, and separate these forced signals from memory. In this talk, I describe a recently developed approach to separate unsteady long-term change from short-term memory, and apply it to surface temperatures to clarify a key point in the scientific debate related to the recent “hiatus” in warming. Given the known ‘ambiguity’ between changepoint models and long-memory models, I will also revisit the surface temperature changepoints detected with long-memory models, and discuss implications for detection and attribution.