Abstarct: In this thesis, we consider optimizing of one of the most important and practical self-organization neural networks, that is, Kohonen neural network using particle filter algorithm. Kohonen neural network has several applications in several areas such as density estimation, image and voice processing, and robotics. In many engineering problems particularly in places that a system has uncertainty or there is a non-Gaussian noise, Monte Carlo methods and especially particle filter have interesting results in compare to other powerful methods such as extended Kalman filter. One serious problem with Kohonen neural network is dead units. This problem, in some cases, leads to a kind of inactivity in network's units, that is, those inactive units do not take part in update stage in algorithm and this may alters the performance of network worst. Various methods have been proposed to solve this problem and in this thesis we proposed a novel extension to correct and optimize the performance of network. Indeed, network's coefficients are estimated by SIR algorithm which by this we could not only solve the dead unit problem, but also we could promote both performance of the network in density estimation and accuracy of the estimation. Preferences of using the particle filter algorithm for learning of Kohonen neural network are high accuracy of Monte Carlo method in estimation and prediction baxsed on restricted data, disuse of feedback to compare the network output to observations, and homogeneous between Kohonen neural network as same as such other unsupervised self-organized neural networks and particle filter algorithm.