The tonguelike curve variable step least mean square (VSLMS) algorithm is a clas sical LMS algorithm. The disadvantage of the algorithm is that its step formula can be disturbed easily by noise jamming, thus causing adaptive filter′s weigh ts fluctuate around the optimal weights. To solve the problem, the tonguelike c urve VSLMS algorithm is improved according to the characteristics that the corre lation of white Gaussian noise is weak. The improved tonguelike curve VSLMS algo rithm based on the correlation characteristic is presented. Tonguelike curve VSL MS algorithm′s ability of anti noise jamming is improved evidently. If two alg orithms choose the same parameters, the improved tonguelike curve VSLMS algorith m based on the correlation characteristic has the less steady state error than the tonguelike curve VSLMS algorithm. Under the condition that the two algorithm s are both convergent, the convergence rate of the improved tonguelike curve VSL MS algorithm is faster than that of the tonguelike curve VSLMS algorithm. The ab ove conclusions are testified through theoretical analysis and simulation.