The method currently used to estimate sugarcane yield in Kenya is made exclusively by a visual physical assessment (VPA). In this method, 15% of the sugarcane fields are sampled and weighted to estimate the total sugarcane production of the following year. This study explores the use of time series of spectral vegetation indices from low resolution satellite images to forecast sugarcane yield in Kenya. Historical yield data (2001 - 2010) for the six sugarcane growing zones in Western Kenya were used to correlate 250 m MODIS-NDVI (MODerate resolution Imaging Spectroradiometer - Normalized Difference Vegetation Index) with historical yield data for multiple growing seasons. When using the whole data set (6 zones and 10 years), data analysis showed that NDVI is neither related to the annual rain, nor to the sugarcane yield. When splitting the dataset according to the year or to the zone, the analysis showed that the relationship between NDVI and yield is generally significant for a given year, and not for a given zone. These results suggest that at the zone scale, the land use is relatively constant nevertheless there is a high land use heterogeneity between zones. This hypothesis was supported by the good correlation that exists between the slope of NDVI and yield at the zone scale, and the fraction of sugarcane crop on the same area (R² = 0.75, p < 0.001). These preliminary results showed that MODIS NDVI can be used to assess yearly sugarcane yield at the zone scale, using historical NDVI and yield data. Next step will consist in characterizing the spatial variability of the zones and sectors, through the development of a satellite-derived "landscape" index both sensitive to the cropping practices and to the environmental conditions. The slope between NDVI and yield, calculated per spatial unit, will be related to this landscape index, in order to develop a generic NDVI-sugarcane relationship that could be applied in Western Kenya.