Some of the most efficient algorithms for finding the discrete logarithm involve pseudo-random implementations of Markov chains, with one or more “walks” proceeding until a collision occurs, i.e. some state is visited a second time. In this paper we develop a method for determining the expected time until the first collision. We use our technique to examine three methods for solving discrete-logarithm problems: Pollard’s Kangaroo, Pollard’s Rho, and a few versions of Gaudry-Schost. For the Kangaroo method we prove new and fairly precise matching upper and lower bounds. For the Rho method we prove the first rigorous non-trivial lower bound, and under a mild assumption show matching upper and lower bounds. Our Gaudry-Schost results are heuristic, but improve on the prior limited understanding of this method. We also give results for parallel versions of these algorithms.