Data mining, which is a technique to extract variable information from enormous data, becomes more and more important. Real data often has missing values. Therefore, a method for estimating the missing data is required in application of data mining. Using multiple self-organizing maps (MSOM) proposed by Kikuchi et al. is one of such estimating method. This method does not need a concrete mathematical model and is also available for nonlinear data. However the performance for various missing patterns were unclear, in addition, the comparisons with conventional imputation methods were not provided. This paper demonstrates the performance and the comparison results through simulation experiments with various missing patterns and conventional methods.