Abstract: It is fundamental for effective thermal management to build an accurate thermal model to predict motor temperature distribution throughout the entire robot. However, quadruped robots contain multiple internal heat sources and complex heat transfer pathways. Existing thermal models focus only on individual motor heat generation, and neglect heat exchange between motors and other heat sources within the compact structure, resulting in reduced accuracy. To address this issue, a whole-body thermal model for quadruped robots is proposed based on the lumped-parameter thermal network, with model parameters identified using the least squares method. This model comprehensively considers motor self-heating, heat exchange among motors and between motors and the onboard computer, heat generated from joint friction and forced convective cooling during locomotion. Temperature distribution predictions are conducted on a quadruped robot platform across various speeds, gaits, and load conditions. The comparison between predicted temperatures at the joint measurement points and actual measurements shows a maximum error of less than 5℃, indicating significantly improved prediction accuracy over thermal models that consider motors individually.