Two identical neutral beam injectors (NBI) have been installed on the COMPASS tokamak. Each of them is capable of producing a deuterium (or hydrogen, helium) neutral beam with nominal power of up to 400 kW at 40 keV. Since the COMPASS tokamak is a compact machine the neutral beams’ design was carefully optimized to fit the neutral beam through the narrow ports of the tokamak. However, once the neutral beams are operated out of the scope of their optimal parameters, additional power losses due to the neutral beam scraping occur. The article investigates the amount of power losses due to neutral beam scraping. The losses are calculated by the neutral beam models and measured neutral beams parameters. The calculation results are compared with the measured power delivered into the plasma. There are two different neutral beam models employed. One describes neutral beam as a sum of small beamlets, which are Gaussian-like. Second one describes neutral beam a single Gaussian beam. The amount of the power lost to scraping is determined by both models and compared. Neutral beam parameters, input parameters for both models, are derived from the Doppler-shifted spectra emitted by fast neutral atoms. It is shown that when operating NBIs at nominal parameters, the scraping losses are negligible (<5% ). If the NBIs are operated to far from the nominal parameters, the scraping losses can reach up to ∼60% of the NBI output power. As the highest fractional losses occur with low power NBI, the beam duct is capable to withstand the corresponding power loads. The expected auxiliary heating power PAUX, the power entering the tokamak chamber carried by fast neutrals, is benchmarked against the additional plasma input power observed on the measured plasma parameters PNBI. The additional plasma input power PNBI is derived from the plasma power balance....