In this paper confidence intervals for the mode based on one observation are constructed. It is proved that there exist distributions for which the length of the confidence Interval can be made in some sense arbitrary close to a non-attainable one. The method for constructing confidence intervals is generalized to multidimensional distributions.