Embodiments of the present application relate to a prediction mode selection method performed at a video encoding device, including: acquiring a first optimal intra-frame prediction mode of a downsampled unit obtained by downsampling an image frame to which a target prediction unit belongs and then dividing the downsampled image frame, and the first optimal intra-frame prediction mode obtained by performing precoding analysis on the downsampled unit; adding a candidate intra-frame prediction mode to a candidate mode set according to the first optimal intra-frame prediction mode; adding a candidate intra-frame prediction mode to the candidate mode set according to a second optimal intra-frame prediction mode of an adjacent PU corresponding to the target PU; and determining an optimal intra-frame prediction mode of a current PU according to prediction residuals and encoding cost values corresponding to the candidate intra-frame prediction modes in the candidate mode set.