The ultimate aim of quantitative remote sensing is inversion. But because of the noise and correlation of observations, the complexity of the land surface system, the information of data is often less than that required to inverse for so many unknowns, the inversion problems of quantitative remote sensing often ill-posed. Though we may have multi-temporal, multi-spectral, and now even multiangular observations, the increasing of information in data is often accompanied with an increasing of number of unknown parameters. Thus in present remote sensing practice, we should make the best use of prior knowledge by a developing suitable inversion strategy. This is the key to resolving this problem. In order to achieve this, we studied the multi-stage, sample-direction dependent, target-decisions (MSDT) inversion strategy. Here a traditional intelligent problem is used to illustrate how prior knowledge is propagating in a multi-stage procedure, and how the information is accumulated in each stage through inversion using new observations. The solution of this problem demonstrates how MSDT concept works. Finally, we discussed how to use information entropy to express the accumulation of our knowledge during the procedure (or MSDT), how to quantitatively describe change of information to help target-decisions.