This paper proposes a model-based and data-driven joint method to estimate the state of health of Li-ion batteries. To accurately quantify battery degradation, a novel resistance-based aging feature is defined from the Thevenin model, and the defined aging feature is approximately linear with capacity degradation. An orthogonal experimental design and a two-way analysis of variance are used to validate the robustness of the defined aging feature. Considering the influence of temperature on battery performance, Box-Cox transformation is introduced to improve the aging feature linearity at low temperatures. Then, an estimator for state of health is established by using Gaussian process regression. Battery aging experiments are conducted to illustrate the estimation effect of the proposed method. The experimental results show that the proposed method has high estimation accuracy at different temperatures. Using the same aging feature, the backpropagation network and support vector regression are implemented to verify the generality of the estimation framework.