We present detailed observations of rivers crossing active normal faults in the Central Apennines, Italy, where excellent constraints exist on the temporal and spatial history of fault movement. We demonstrate that rivers with drainage areas N 10 km 2 and crossing faults that have undergone an increase in throw rate within the last 1 My, have significant long-profile convexities. In contrast, channels that cross faults that have had a constant-slip rate for 3 My have concave-up profiles and have similar concavities and steepness indices to rivers that do not cross any active fault structures. This trend is consistent across the Central Apennines and cannot be explained by appeal to lithology or regional base level change. The data challenge the belief that active faulting must always be reflected in river profiles; instead, the long-profile convexities are best explained as a transient response of the river system to a change in tectonic uplift rate. Moreover, for these rivers we demonstrate that the height of the profile convexity, as measured from the fault, scales with the magnitude of the uplift rate increase on the fault; and we establish that this relationship holds for throw rate variation along strike for the same fault segment, as well as between faults. These findings are shown to be consistent with predictions of channel response to changing uplift rate rates using a detachment-limited fluvial erosion model, and they illustrate that analysis of the magnitude of profile convexities has considerable predictive potential for extracting tectonic information. We also demonstrate that the migration rate of the profile convexities varies from 1.5-10 mm/y, and is a function of the slip rate increase as well as the drainage area. This is consistent with n N 1 for the slope exponent in a classical detachment-limited stream-power erosion law, but could potentially be explained by incorporating an erosion threshold or an explicit role for sediment in enhancing erosion rates. Finally, we show that for rivers in extensional settings, where the response times to tectonic perturbation are long (in this case N 1 My), attempts to extract tectonic uplift rates from normalised steepness indices are likely to be flawed because topographic steady state has not yet been achieved.