We investigate an alternative way of constructing a confidence interval based on M-estimator for a single parameter in a linear model. We compare this confidence interval with a traditional (Wald type) confidence interval theoretically as well as by the means of a Monte-Carlo experiment.