The frequency function of a linear dynamical system can be estimated using spectral estimation techniques. Traditionally, this is done by smoothing in the frequency domain or windowing in the lag domain. These methods are successful in most cases but sometimes they yield estimates that suffer from poor resolution. Typically, this happens when fine details are estimated. In this paper an alternative method is proposed. The method combines traditional windowing with results in adaptive time-frequency analysis. In simulations concerning systems which are difficult for the traditional approach, this method gives better better estimates of the frequency function.