One of the important tasks in the spike train analysis is to estimate the underlying firing rate function. The aim of this article is to improve the time performance of an algorithm which can be used for the estimation. As there is no unique way how to infer the firing rate function, several different methods have been proposed. A popular method how to estimate this function is the convolution of the spike train with Gaussian kernel with appropriate kernel bandwidth. The definition of what “appropriate” means remains a matter of discussion and a recent paper [1] proposes a method how to exactly compute optimal bandwidth under certain conditions. For large sets of spike train data the elementary version of the algorithm is unfortunately too inefficient in terms of computational time complexity. We present a refined version of the algorithm which in turn allows us to use the original method even for large data sets. The achieved performance improvement is demonstrated on a particular results and shows usability of proposed method.