We created virtual soft media by using a haptic device and performed a sensory test focusing on proprioceptive sensation in muscles and tendons. The depths of the virtual media were 30 mm and their spring rates changed according to the displacement. Subjects handled the end-effecter of the haptic device and pushed a pointer into virtual media in a display monitor to sense feedback force calculated by a personal computer. Two experiments were performed to elucidate a part of the sensory perception mechanism. In the experiment 1, subjects evaluated the hardness of the virtual media sensing the whole mechanical characteristics from the surface to the bottom. In the experiment 2, they evaluated the hardness in the range from the surface to the depth of just over 25 mm. The hardness evaluation scores in the exp. 2 were significantly higher than those in the exp. 1 on ANOVA. The changes of spring rates also affected the hardness evaluation scores significantly. We examined the optimal model, which explained the sensory perception mechanism, with covariance structure analysis. As a result, the optimal models of the exp. 1 and the exp. 2 were unified, and the hardness evaluation scores were explained with the mean values and the slopes of logarithmic spring rates. This optical model suggested that subjects could correctly acquire the hardness information of soft media by using the mean values and the slopes, even though they did not have the whole mechanical characteristics.