Résumé: In aquaculture, it is extremely important to determine the composition of fish farm waste to evaluate its potential impacts and to improve its reuse. Near-infra red spectroscopy (NIRS), an alternative to standard chemical analytical techniques, is a quick non-invasive method to assess physical and chemical composition, reducing the cost of routine analysis. We developed NIRS calibration models for organic matter (OM), total organic carbon (TOC), total organic nitrogen (TON), the carbon/nitrogen ratio (C/N), total phosphorus (TP) and the lipid content of marine fish particulate waste. To obtain a wide range of compositions of fish waste, decomposition time, feed loss, and inter-specific variations were taken into account. The NIRS calibration models were built using three sub-datasets: in Scenario 1) the calibration was species-specific, including only seabass waste (SeabassWaste), in Scenario 2), the calibration included data from two other species (MultiSpeciesWaste) and in Scenario 3), the general calibration included all data as well as simulation of extreme feed loss (up to 50%) (Faeces&Feed). All calibrations performed using either dried or wet samples gave equations with high coefficients of determination (R2) and reasonably low standard error of cross validation (SECV) values for all parameters tested, except for TP due the high proportion of mineral P.