In order to improve the stability and accuracy of the object tracking under nonlinear conditions,an object tracking algorithm based on partial least square (PLS) representation and classification joint optimization is proposed.In this method,object tracking is defined as an optimization task to m inimize the representation error and classification loss.Firstly,PLS theory is used to express foreground informa tion,and multiple linear appearance models are constructed to describe the no nlinear appearance changes of the object. Then,a joint optimization objective function based on deterministic search mech anism is proposed.Furthermore,a stochastic gradient classifier (SGC) based on in cremental feature updating is proposed,which makes sure to achieve a stable trac king.Experiments show favorable performance of our algorit hm on sequences where the targets undergo various complex changes on for eground and background.