A beam training method and a related device are disclosed. The method includes: sending, by a first communications device, sidelink control information (SCI) to a second communications device, where the SCI is used to instruct the first communications device to send at least one channel state information-reference signal (CSI-RS) to the second communications device; and sending, by the first communications device, the at least one CSI-RS to the second communications device based on the SCI, where the at least one CSI-RS is used by the second communications device to determine that a beam corresponding to a CSI-RS with highest signal strength in the at least one CSI-RS is a beam for communication between the second communications device and the first communications device. The CSI-RS used for beam training is dynamically sent based on the instruction of the SCI.