Joint stiffness is defined as the dynamic relationship between the position of the joint and torque acting about it Joint stiffness is composed of two components: intrinsic and reflex stiffness. Measuring the two stiffness components cannot be done simply because the two components appear and change together. A number of approaches have been used to estimate the components, but all those approaches are inherently off-line. We have developed a novel algorithm that separates and estimates the two components in real-time. Intrinsic stiffness was estimated by finding the cross-correlations between the position, its derivatives and the torque. Reflex stiffness was estimated by finding the IRF between the half-wave rectified velocity and the estimated reflex torque. A novel position perturbation, consisting of pseudo random series of pulses of different lengths, was used to eliminate covariance of intrinsic and reflex stiffness estimates. Using simulated data, the real-time estimates were shown to be estimated accurately. The real-time estimation algorithm was validated by comparing the real-time estimates with estimates generated by the parallel-cascade identification, an established off-line intrinsic and reflex stiffness identification algorithm, using simulated and experimental data. The estimates produced by the two algorithms were in agreement for both simulated and experimental data.