SPH methods are used for simulating the waves interaction on coastal defence structures. The Navier-Stokes equations are discretized by means of a particular representation. The particle approximation evaluates the physical quantities and their gradient using a kernel function [1][2]. The kernel function impacts the accuracy and stability of SPH methods due to the tensile instability phenomenon [3]. This phenomenon is related to the pressure gradient term. The expansion of this term shows the influence of the pressure sign on the Navier-Stokes equations, an attraction or repulsion of particles. The introduction of an artificial force term in the Navier-Stokes equations was proposed to correct this particle distribution problem [4]. The choice of the kernel function is another possible way to circumvent the problem [5]. In this work, several kernel functions are studied in order to select the one suited for free surface flows. The dam break benchmark, a case with important free surface strains, will be used. The characterization of suited kernel functions for our problem will be based on the standard deviation used to express the smoothing error criterion [6]; the kernel function is chosen according to its ability to minimize the standard deviation, to deal with the tensile instability problem mentioned earlier, and to reproduce the experimental results. Multiple kernel functions were tested in our study, from different categories. We retain four of them, for which the results are presented in Fig. 1. and Fig. 2. In our results, the Double Cosine kernel function with k=2 stands out, which is coherent with our calculations of the standard deviation, namely σ=0.5 for the Gaussian in [0,∞], σ=0.32 for the Double cosine with k=2, σ=0.33 for the Cubic and σ=0.38 for the Wendland, in their compact support.