The return of portfolios will be limited when one uses the variance of portfolio to control risk． Compared with the shortcoming of the variance，lower partial moment (LPM) is paid close attention by many researchers due to the fact that LPM controls only the risk of portfolios without limiting the return． But，under the assumption of non-normal distribution，one can not generally obtain the analytic properties of LPM models． Motivated by these facts，under the assumption of uncertainty distribution，we propose a class of robust tracking error portfolio selection problems with multiple weights constraints in which we use the worst-case lower partial moment (LPM) to measure the loss of portfolios． The analytic solutions of the proposed robust models with-order LPM constraints are obtained． Some interesting and novel results are found based on the geometrical presentations of efficient frontiers． The numerical comparisons indicate that the proposed robust models have much better performance than the classical mean-variance model．