Abstract Particle-in-cell (PIC) models have been extensively used in the last few years to describe negative ion extraction for neutral beam injection applications. We show that some of these models have been employed in conditions far from the requirements of particle simulations and that questionable conclusions about negative ion extraction, not supported by experimental evidence, have been obtained. We present a critical analysis of the method that has led to these conclusions and propose directions toward a more accurate and realistic description of negative ion extraction. We show in particular that, as expected in PIC simulations, mesh convergence is reached only if the grid spacing is on the order of or smaller than the minimum Debye length in the simulation domain, and that strong aberrations in the extracted beam are observed if this constraint is not respected. The method of injection of charged particles in the simulated plasma is also discussed, and we show that some injection methods used in the literature lead to unphysical results.