We develop an empirical tool to estimate optimal highway durability in cold regions. To test the model, we assemble a data set containing all highway construction and maintenance projects in Arizona and Washington State from 1990 to 2014. The data set includes information on location, time, type (resurfacing, construction, or lane widening), pavement material and thickness, and total expenditure for these projects. Using the data, we first estimate how highway maintenance costs and highway duration depend on pavement thickness and traffic loading. We then calibrate the effects of different deicers on highway durability and thus on highway maintenance costs. Finally, we demonstrate how the estimated and calibrated model can be used by planners to make optimal decisions for highway pavement and winter operations in cold regions.