Wave energy converter power production assessment, usually carried out using a power matrix, is essential for the appraisal of new wave energy converter technologies and for the planning of specific wave energy projects. Errors in power assessment may arise, both from an inaccurate description of the wave energy converter dynamics and from an excessively simplified representation of wave spectra in the power matrix approach. Ideally, the wave energy converter output should be computed in every individual sea state of the wave dataset considered, without the assumption of any parametric spectral shape. However, computationally efficient methods are necessary to achieve such extensive wave energy converter simulation. The non-linear frequency-domain technique is significantly faster than Runge–Kutta time-domain simulations, without affecting the representation of radiation forces and non-linear dynamics. In this article, the two main sources of errors in wave energy converter power assessment, namely the power matrix representation and wave energy converter modelling inaccuracies, are jointly studied and put into perspective, using four case studies (two wave energy converter systems in two locations). It is found that both types of errors can be of comparable magnitude. The non-linear frequency-domain technique simulation technique is shown to be a computationally efficient tool, retaining a realistic representation of the device dynamics while avoiding the use of a power matrix, thus preserving accurate representation of both the sea states and the wave energy converter, at little computational expense. Aside from those main results, the issue of the length and number of simulations, necessary to achieve average power estimates with sufficient accuracy in every sea state, is addressed in detail.