Local abundance and dynamics of sedentary species with a dispersing life stage reflect factors that influence input and loss rates to patches of suitable habitat. For reef fishes, more attention has focused on sources of variation in input (larval settlement) than on patterns and causes of subsequent losses. We estimated spatial and temporal variation in juvenile mortality of a tropical damselfish, yellow-tail dascyllus (Dascyllus flavicaudus; Pomacentridae), using a fixed density experiment that was repeated 5 times at the same eight mid-lagoon localities at Moorea, French Polynesia. There was little temporal variation in the overall percent of outplanted fish lost in 48 h among five time periods (range: 32-37%), whereas there was substantial variation among the sites in the average percent lost (range: 16-56%). Differences in loss rates among the sites were highly consistent among the time periods. Densities of predators of juvenile dascyllus varied substantially among the eight sites and were highly correlated with loss rate of dascyllus. We used the empirically derived relationship between predator density and damselfish loss rate to predict the loss rate of dascyllus at four additional sites, and there was excellent agreement between the predicted and observed loss rates. There was a strong positive relationship between predator densities at the 12 sites and structural attributes of the reefs that do not change on a fast time scale, suggesting why there was strong spatial and weak temporal variation in mortality rates, with no interaction between spatial and temporal variation. Natural settlement rates of yellow-tail dascyllus and of a close congener (humbug dascyllus, D. aruanus) varied among the sites, and settlement of the two species was inversely correlated (r=-0.68). Settlement of these species was not statistically correlated with variation in mortality rate, but there was a weak trend for settlement of yellow-tail dascyllus to be greater at sites with higher mortality (r=0.27), and for settlement of humbug dascyllus to be greater at sites with lower mortality (r=-0.32). We calculated that even these weak co-variances could reduce (yellow-tail dascyllus) or increase (humbug dascyllus) the spatial variance in density of 48-h-old recruits arising at settlement by 19 and 27% respectively. Taken together, the findings suggest that the interactions between and relative contributions of input and loss processes can differ substantially over a scale of a few kilometers, resulting in a mosaic of local patches characterized by different abundances and dynamics.