Temperature variations have significant effects on guide wave propagation and therefore increase the detection uncertainty of the guided wave–based structural health monitoring system. A novel temperature compensation technique combining an adaptive filter and optimal baseline selection is developed to enhance the robustness and effectiveness of guided wave–based damage detection. The adaptive filter is the finite length unit impulse response digital filter based on adaptive linear neuron network. This article focuses on three main issues for practically implementing the proposed method: (a) establishment of temperature compensation standard, (b) parameter design of compensation filter, and (c) determination of temperature gradient to reduce the number of selected baselines. Experiments are conducted on two stiffened composite plates to verify the proposed method for effective and robust temperature compensation under a large temperature range from −40°C to 80°C. Results show that temperature interval for baselines of low-frequency signals, such as 50 kHz, can be up to 20°C to provide good temperature compensation with the proposed method, while temperature interval for baselines of high-frequency signals, such as 450 kHz, can be up to 12°C.