Using Bayesian statistical models, the authors investigated spatial and temporal variations in small-area malaria incidence rates for the period mid-1986 to mid-1999 for two districts in northern KwaZulu Natal, South Africa. Maps of spatially smoothed incidence rates at different time points and spatially smoothed time trends in incidence gave a visual impression of the highest increase in incidence occurring where incidence rates previously had been lowest. This was confirmed by conditional autoregressive models, which showed that there was a significant negative association between time trends and smoothed baseline incidence before the steady rise in caseloads began. Growth rates also appeared to be higher in the areas close to the Mozambican border. The main findings of this analysis were that: 1) the spatial distribution of the rise in malaria incidence is uneven and strongly suggests a geographic expansion of high-malaria-risk areas; 2) there is evidence of a stabilization of incidence in areas that had the highest rates before the current escalation of rates began; and 3) areas immediately adjoining the Mozambican border appear to have undergone larger increases in incidence, in contrast to the general pattern of low growth in the more northern, high-baseline-incidence areas, but this was not confirmed by modeling. Smoothing of small-area maps of incidence and growth in incidence (trend) is important for interpretation of the spatial distribution of disease incidence and the spatial distribution of rapid changes in disease incidence.