Accurate diagnosis of tree health is the foundation of urban forest management as well as a technology urgently needed in the production. Tree health diagnosis through the analysis of the soil and plant nutrient health shows poor reliability while morphological diagnosis appears time-consuming and laborious, so how to diagnose the tree health fast, accurately and nondestructively has become an important technical bottlenecks of urban trees health management. This paper, took ginkgo trees in Beijing as the research object, has studied the trees health diagnosis technology based on the leaf spectral reflectance characteristics. Based on the clustering analysis of 13 exterior shape features, trees were divided to four healthy levels, excellent, good, fair, and poor. Leaf pigment content between different healthy levels of trees is extremely significant different (p<0.001). Because there is relationship between the spectral reflectance and chlorophyll content, the tree leaf spectral reflectance characteristics used for health diagnosis is feasible. Adopted the factor analysis, we constructed green degree index, index of pigment, trilateral index, reflecting leaf spectral reflectance characteristics, according to 15 leaf spectral reflectance indices. Leaf spectral reflectance indexes and three reflection spectrum indexes between different healthy levels of trees had extremely significant difference (p<0.001). Three reflection spectrum indexes were used to construct the multiple quadratic model of the ginkgo trees health evaluation, with prediction accuracy of up to 79%. Therefore, this method can be used as a rapid diagnosis method to evaluate ginkgo trees health. This study selected relatively comprehensive and concise spectral indexes, determined the comprehensive score and score range for the tree core morphological index, green degree index, index of pigment, trilateral index of leaves of different healthy levels of trees. It provides a standard for the health diagnosis of ginkgo trees in practical forest management.