Abstract: New Caledonian crows Corvus moneduloides (NC crows) display sophisticated tool manufacture in the wild, but the cognitive strategy underlying these skills is poorly understood. Here, we investigate what strategy two free-living NC crows used in response to a tool-length task. The crows manufactured tools to extract food from vertical holes of different depths. The first tools they made in visits were of a similar length regardless of the hole depth. The typical length was usually too short to extract food from the deep holes, which ruled out a strategy of immediate causal inference on the first attempt in a trial. When the first tool failed, the crows made second tools significantly longer than the unsuccessful first tools. There was no evidence that the crows made the lengths of first tools to directly match hole depth. We argue that NC crows may generally use a two-stage heuristic strategy to solve tool problems and that performance on the first attempt in a trial is not necessarily the 'gold standard' for assessing folk physics.