Abstract:There is a complicated relation between thin-bed thickness and the corresponding seismic reflection characteristics,and the relation can be described by usingthree layer .neural network. We can derive thinbed thickness from seismic data by inputting into the network 6 seismic characteristic values (the maximum amplitude value in time window,the maximum amplitude-spectrum value,frequency corresponding to the maximum amplitude-spectrum value,the ratio of the maximum autocorrelationfunction value to the minimum one,the center frequency and low-frequency energy) , and then by making appropriate conversions of both the input values and the output values in the neural network, In trial modeling computation,theneural network recommended here can be used to estimate correct thin-bed thickness when the signal/noise ratio of seismic data is high enough. But when the seismic signal/noise ratio is low,we can correctly estimate a thin-bed thickness by using the neural network trained with the use of the sample trace that has higher signal/noise ratio. This method has been used in real data processing to bring satisfactory result.