This study looked at how well commonly used climate datasets, and methods of projecting future change, capture local temperature patterns in the Puget Sound area of Washington State. We found that high-resolution historical climate data can contain biases, especially in places with complex terrain like mountains or coastlines, which may not be apparent to users. These errors can then carry over into climate projections used for planning the future. When we compared three widely used methods for refining global climate models into high-resolution projections, we found substantial differences between datasets—sometimes as much as 50 days per year in how often freezing temperatures were projected, or over 1 degree Celsius in projected winter warming. This means that depending on which dataset or method is used, very different outcomes might be expected for the same location. Our findings show that just because climate data is high-resolution doesn’t mean it’s necessarily accurate at the point scale, and decision-makers should be aware of the limits of these tools when planning for climate risks.