Software engineering society has always faced the problems of accuracy of Software effort estimation. To advance the estimation accuracy of software effort, many studies have focused on effort estimation methods without any concern of data quality, although data quality is one of important factor to impact to the estimation accuracy. So I investigated the influence of outlier elimination upon the accuracy of software effort estimation through experiments applying two outlier elimination methods (K-means clustering and My-K-means clustering) and two effort estimation methods( Least squares and Neural network) associatively. A new outlier elimination method My-K-means clustering is proposed which gives better estimation results than K-means clustering. The experiments were performed using the Bank data set which consists of the project data performed in a bank in Pakistan, with or without outlier elimination.