We present new techniques for computing exact p-values of Kendall's rank correlation statistic when ties are present. An explicit formula for the probability-generating function of this statistic in the case of ties in one ranking is derived. This allows for computation of exact p-values for sample sizes up to 100, regardless of the tie structure. When ties are present in both rankings, one has to fall back on enumeration methods. We discuss how a large part of this enumeration can be avoided. The exact results for ties in one ranking are used to develop an efficient Monte Carlo simulation algorithm for approximating p-values in case of ties in both rankings. Finally, we shortly discuss computer implementations of our methods, which are available on the internet.