The length of the solar cycle and its long-term variation have recently received additional significance due to their suggested connection to global climate. The cycle length is conventionally defined as the time difference between two successive sunspot minima. However, the sunspot minimum times sensitively depend on the way the sunspot numbers are averaged, i.e., whether one uses daily or monthly averaged values and whether and how the data are further smoothed. Using differently processed sunspot data, the sunspot minimum times vary typically by a few months leading to a corresponding inaccuracy in solar cycle length. Here we propose a new method to define the solar cycle length as a difference between the median activity times of two successive sunspot cycles. The great advantage of this method is that the median times are almost independent of how the sunspot minima are determined. Therefore the method allows the solar cycle lengths to be calculated with a very small inaccuracy of a few days only. We show that the individual cycle lengths calculated from the conventional and the median method may differ by nearly a year. However, the long-term trend of cycle lengths remains roughly the same during modern times. ¿ 1998 American Geophysical Union