The paper proposes a method for estimating strength reliability of machine parts based on computer modeling and the use of nonparametric statistics methods, which allows taking into account any laws of part loading: both standard and random, set by an operational sample. All algorithms are implemented in the MathCad mathematical processor, the Parzen-Rosenblatt estimate is used to restore the unknown distribution density function of random variables. Using examples of calculations of gears, it is shown that when a gear operating condition changes from light to heavy, the failure probability due to contact strength, changes by more than an order of magnitude: the failure probability of the 2207 rolling bearing in terms of dynamic load capacity by 3.6 times, the shaft failure probability by almost 20 times. The results of calculating the gear failure for a bimodal loading condition are presented, even an approximate estimate of which is impossible by classical methods. The verification of the developed methodology based on a calculation of the locomotive bogie frame in a mean normal loading condition gives a failure probability of 9.44% which is in good agreement with the frame failure figures: 10%. The proposed method of probabilistic calculation of parts is universal. If the loading condition is set by a sample or other distribution, it is implemented unchanged.