Given the effectiveness of semiconductor devices for classical computation one is naturally led to consider semiconductor systems for solid state quantum information processing. Semiconductors are particularly suitable where local control of electric fields and charge transport are required. Conventional semiconductor electronics is built upon these capabilities and has demonstrated scaling to large complicated arrays of interconnected devices. However, the requirements for a quantum computer are very different from those for classical computation, and it is not immediately obvious how best to build one in a semiconductor. One possible approach is to use spins as qubits: of nuclei, of electrons, or both in combination. Long qubit coherence times are a prerequisite for quantum computing, and in this paper we will discuss measurements of spin coherence in silicon. The results are encouraging - both electrons bound to donors and the donor nuclei exhibit low decoherence under the right circumstances. Doped silicon thus appears to pass the first test on the road to a quantum computer. Comment: Submitted to J Cond Matter on Nov 15th, 2005