Existing fault-tolerant clock synchronization protocols are shown to result from refining a single clock synchronization paradigm. In that paradigm, a reliable time source periodically issues messages that cause processors to resynchronize their clocks. The reliable time source is approximated by reading all clocks in the system and using a convergence function to compute a fault-tolerant average of the values read. The performance of a clock synchronization algorithm based on the paradigm can be quantified in terms of the two parameters that characterize the behavior of the convergence function used: accuracy and precision.