One of the most important applications of hyperspectral data analysis is either supervised or unsupervised classification for land cover mapping. Among different unsupervised methods, partitional clustering has attracted a lot of attention, due to its performance and efficient computational time. The success of partitional clustering of hyperspectral data is, indeed, a function of five parameters: 1) the number of clusters, 2) the position of clusters, 3) the number of bands, 4) the spectral position of bands, and 5) the similarity measure. As a result, partitional clustering can be considered as an optimization problem whose goal is to find the optimal values for above-mentioned parameters. Depending on this fact that which of these five parameters entered to the optimization four different scenarios have been considered in this paper to be resolved by particle swarm optimization. Our goal is, then, finding the solution leading to the best accuracy. It should be noted that among five different parameters of clustering, both similarity measure and the number of clusters have been considered fixed to prevent over-parameterization phenomenon. Investigations on a simulated dataset and two real hyperspectral data showed that the case in which the number of bands has been reduced in a pre-processing stage using either band clustering in the data space or PCA in the feature space, can result in the highest accuracy and efficiency for thematic mapping.