Walking speed (WS) predicts mortality. However, it is unclear if decline in WS increases prior to death. We examined whether (a) WS declined faster in persons who died during the follow-up compared with those who remained alive and (b) adding change in WS to a model including age, sex, and baseline WS improved prediction of mortality. Data are from 4,016 participants of the Dijon center of the Three-City study (France), aged 65-85 years. Fast WS (FWS) was measured up to five times over a 12-year period. Mortality was ascertained until 2012. Linear mixed models using a backward time scale showed that FWS declined faster in 908 participants who died during the follow-up (annual change = -0.031 m/s) than in those who survived (-0.021 m/s), corresponding to a difference of -0.009 (95% confidence interval = -0.013 to -0.005) m/s. Compared with "normal" change in FWS (annual change ≥-0.04 m/s), "substantial" decline (<-0.08 m/s) was associated with a 1.4-fold greater risk of mortality (hazards ratio = 1.40, confidence interval = 1.02-1.92) and small decline (-0.08 to -0.04 m/s) with a 1.2-fold greater risk (hazards ratio = 1.18, confidence interval = 0.89-1.57). The net reclassification index when adding these categories of change in FWS to the model adjusted for age, sex, and baseline FWS was 19.0% (0.6, 36.8%). Participants who died during the follow-up had a steeper decline in FWS than the others. Both baseline FWS and FWS decline predict mortality.