Recorded signals from a guided wave structural health monitoring system are sensitive to damage but also to even small temperature changes. Previous work has shown that reasonable temperature compensation can be achieved by first selecting an optimal baseline and then stretching it to match the current signal of interest. However, this method is not perfect and its efficacy depends upon many factors. This paper considers each step of this existing method and implements modifications to improve the efficacy of temperature compensation. The criterion for evaluation is a comparison of the residual signals after baseline subtraction before and after damage is introduced. Results show that damage detection performance is significantly improved with this new method.