 This paper proposes an alternative approach to creating seasonal reflectance images from multi-temporal satellite data. Instead of selecting the maximum value of normalized difference vegetation index, NDVI, for each pixel across all available dates, the authors select the medoid, a multi-dimensional analog of the median, for each pixel. This results in a single observation for each pixel, allowing for the preservation of relationships between bands. The method was tested on Landsat TM slash ETM plus imagery, and the resulting seasonal reflectance images were found to be more representative of the time series than the maximum NDVI seasonal composite. This article was authored by Neil Flood.