The Monte Carlo simulation (MCS) is an established tool for probabilistic analyses. It allows to estimate statistics of output values, to obtain sensitivities, and to build surrogate models. The quality of these results is increased with increasing sample size, however, this comes with additional computational cost. In practice, a compromise must be found between accuracy and time. Hence, the goal is to extract as much information as possible with one single sample. In this paper, Latinized Particle Sampling (LPS) is introduced as a new sampling method, which distributes the samples uniformly in the sample space. For this, the samples are considered as charged particles, which repel each other. In an iterative process a force equilibrium is obtained. In order to obtain the desired marginal distributions, the sample is latinized, giving a valid Latin Hypercube design. Additionally, a correlation control algorithm is applied to obtain a desired target correlation. Due to the uniform space filling, the quality of surrogate models is increased in comparison to regular Latin Hypercube sampling (LHS). Compared to an optimized LHS design, the surrogate model quality of LPS are lower, but LPS samples can be created much faster.