- Today the most popular method for the extraction of vegetation information from remote sensing data is through vegetation indices. In particular, erosion models are based on vegetation indices that are used to estimate the cover factor (C) defined by healthy, dry, or dead vegetation in a popular soil erosion model named RUSLE, (Revised Universal Soil Loss Equation). Several works correlate vegetation indices with C in order to characterize a broad area. However, the results are in general not suitable because most indices focus only on healthy vegetation. The aim of this study is to devise a new approach that automatically creates vegetation indices that include dry and dead plants besides healthy vegetation. For this task we propose a novel methodology based on Genetic Programming (GP) as summarized below. First, the problem is posed as a search problem where the objective is to find the index that correlates best with on field C factor data. Then, new indices are built using GP working on a set of numerical operators and bands until the best composite index is found. In this way, GP was able to develop several new indices that are better correlated compared to traditional indices such as NDVI and SAVI family. It is concluded with a real world example that it is viable to synthesize indices that are optimally correlated with the C factor using this methodology. This gives us confidence that the method could be applied in soil erosion assessment. © Springer-Verlag Berlin Heidelberg 2009.