Growth variability in pearl millet (Pennisetum glaucum) over short distances is a severe constraint on the interpretation of agricultural experiments in the West African Sahel. The purpose of this study, therefore, was to compare different non-destructive methods to estimate, spatially, millet growth and final yields. Aerial photography, georeferenced radiometric measurements and a chlorophyll meter were tested during three rainy seasons in a nitrogen rate×density×genotype experiment in western Niger. For the radiometric measurements, normalized difference vegetation indices (NDVI) obtained and calibrated for individual millet hills spaced 1.5 m apart were aggregated for the entire experiment with 6000 samples per hectare. A simple calibration procedure was used to correct for variation in soil background reflectance and incident light. For NDVI measurements of individual planting hills, the correlation between plant total dry matter (TDM), leaf weight, leaf area and NDVI was high (r2 = 0.89–0.91) and regression parameters were genotype-specific. Aggregated georeferenced NDVI measurements at the plot level correlated with grain and TDM at harvest (r2 = 0.40–0.87). The analysis of true-colour and infrared aerial photographs permitted the monitoring of millet growth and the quantitative evaluation of treatment responses throughout the growing season. The infrared images were the most efficient in the detection of vegetation followed by the normalized green band of true-colour images. The red band was the least effective because of the influence of soil albedo and image vignetting. Although chlorophyll meter measurements reflected relative differences in plant nitrogen status between treatments, their interpretation required destructive sampling and proved unsuitable to predict millet yields. The results demonstrate the potential of georeferenced radiometric data and aerial photographs to improve soil sampling strategies, sequential plant growth monitoring and the statistical design and analysis of experiments. By providing intermediate data sets, the tested tools can also help in the upscaling of ground truth to satellite data in yield prediction studies.