The relationships between satellite-viewed cloudy (or partly cloudy) grid cells and the variability of the precipitation contained therein are explored. Using a 32 km grid and 30 min interval visible, infrared and radar data, 5 days of the Florida Area Cumulus Experiment are examined. Cloud is delineated from no-cloud by an infrared threshold of 253 K. While high rainrates are always associated with low temperatures, the reverse is not true: low temperatures do not always imply high rainrates. For partly cloudy cells, the percent-explained variance of rainrate by infrared parameters is low, with none of the parameters explaining more than 14% of the variance. The mean visible count explains slightly more variance, but it is not apparent that higher visible values are indicative of higher rainrates, because the higher resolution of those data introduces ground pixels into the average. When only completely cloudy cells are considered, the infrared parameters still explained about 14% of the variance, but with larger day-to-day variability. For those cells, the mean visible count explains less than 10% variance on 4 of the 5 days, due to its inability to discern rainrates in widespread cirrus anvils. The mean visible structure by itself explains 10%–26% of the rainrate variance for completely cloudy grid cells. Modest (4%–14%) increases in explained variance are shown when this quantity is then added as a second regression parameter. Classification of the mean rainrate into six groups and the subsequent computation of a mean infrared parameter for each class shows statistically significant differences in the mean infrared parameters among classes. Assigning independent observations to classes becomes unsatisfactory given the distribution of the rain classes themselves. Variability (between days) in the mean temperature of each rainrate class is often as great as the variability (of the mean temperature) among rain classes on any given day. Relationships are clearly dependent on where in the convective cycle they occur, and this cycle is itself variable from day to day. Extensive cold anvils often produce widespread stratiform rain late in the day, while earlier these same temperatures produced intense convective rain. On the scales examined here, the results indicate that useful, accurate rainfall estimates beyond rain/no-rain discrimination are unlikely.