A novel NLP task, automatic survey coding, is described, and two methods for performing this task are presented. The first method uses a Boolean pattern-matching strategy to code survey responses, while the second uses a vector-based (probabilistic) method. The performance of the two methods is tested and compared on three representative survey datasets. The Boolean method is shown to perform slightly better on average than the vector-based method. Linguistic factors affecting the difficulty of the coding task for each survey are discussed.