In this paper the problem of non-parametric estimation of the probability density function for hydrological data is considered. For a given random sample X1, X2, ..., Xn we define an estimator fˆn of the density function ƒ based on a function K of a real variable – the so-called kernel of a distribution – and a properly chosen number sequence {hn} from the interval (0, ∞). This estimator of density function of a random variable X under more general assumptions is known in the statistical literature as the Parzen-Rosenblatt estimator or the kernel estimator. The method of kernel estimation presented in the paper has been applied to determine the probability distribution of the groundwater level based on long-term measurements made in the melioration research carried out at the foothill object Długopole.