A survey on the quality of poultry feeds was run in 7 east African countries (Kenya, Tanzania, Uganda, Ethiopia, Sudan, Eritrea, Burundi). The objective was to provide a reliable overview of poultry feeds available on the local markets in order to advise farmers for feed use, and help feed millers to improve quality of their products. Approximately 400 samples were collected across the countries in order to cover the variability of all commercial feeds available. They corresponded to broiler and layer feeds, and originated from various suppliers and resellers. Chemical analysis of samples was performed on 150-200 samples selected for their spectral representativeness of the whole database (clustering based on neighbourhood distance calculated from principal components). The parameters measured included Dry Matter (DM), Total Minerals (MM), Crude Protein (CP), Crude Fat (CFat), Crude Fibre (CFib) and Starch (ST). The Metabolizable Energy (ME) value was calculated from chemical composition. NIR analysis was performed on the samples used for chemical analyses (grinding with 1 mm sieve). Spectra in diffuse reflectance were measured on a FOSS NIRSystem 6500 spectrometer in small ring cups. Calibration equations were built after mathematical pre-processing of data (SNV and detrend with 2nd derivative of spectra) without visible wavelengths. Partial Least Squares regression (modified PLS in WINISI software) was found to be the most efficient method for calibrations. Equations obtained had SECV (SE of cross-validation) values of 0.43%, 2.0%, 0.61%, 0.39%, 0.43%, 1.3% and 81.3kcal/kg for DM, MM, CP, CFat, CFib, ST and ME respectively. Except for minerals, the performance of equations was satisfactory and allowed an efficient quality control. They were therefore used for the prediction of the chemical composition of samples not analysed with reference methods. It is concluded that NIRS analysis can be efficient to predict chemical composition of poultry feed in east Africa. T