Abstract: The sandstone reservoir in Dongpu depression is deep and thin,changes fast laterally,conventional seismic data can't finely describe the reservoir characteristics.Seismic reservoir prediction technologies such as spectrum inversion,frequency division interpretation,simulated annealing inversion and multi-attributes must be used to study the thin reservoir characteristics.The first process is to use spectrum inversion and frequency division interpretation technology,transform time domain seismic data into frequency domain,the second step is to use simulated annealing inversion to confirm the thickness of thin sand,then to use multi-attributes such as wave impedance,frequency and arc length and so on to predict the sand distribution ranges.They were used synthetically to reduce the multi-solution of prediction results.The results offered us the accurate spatial distribution of thin sand reservoir filled with oil and gas.The prediction results were verified by drilling and could guide the further exploration.