Online forums or message boards are rich knowledge-based communities. In these communities, thread retrieval is an essential tool facilitating information access. However, the issue on thread search is how to combine evidence from text units(messages) to estimate thread relevance. In this paper, we first rank a list of messages, then we score threads by aggregating their ranked messages' scores. To aggregate the message scores, we adopt several voting techniques that have been applied in ranking aggregates tasks such as blog distillation and expert finding. The experimental result shows that many voting techniques should be preferred over a baseline that treats a thread as a concatenation of its message texts.