This paper presents the knee-joint vibration signal processing and pathological localization procedures using the empirical mode decomposition for patients with chondromalacia patellae. The artifacts of baseline wander and random noise were identified in the decomposed monotonic trend and intrinsic mode functions (IMF) using the modeling method of probability density function and the confidence limit criterion. Then, the fluctuation parts in the signal were detected by the signal method ting for count. The results demonstrated that the quality of reconstructed signal can be greatly improved, with the removal of the baseline wander (adaptive trend) and the Gaussian distributed random noise. By detecting the t signals in the artifact-free signal, the pathological segments related to chondromalacia patellae can be effectively localized with the beginning and ending points of the span of t signals.