In ultrasonic NDE of materials, deconvolution techniques are widely used to improve time/space resolution, minimize spectral coloring, and compensate for different experimental settings, e.g., transducer variations, pulser-receiver energy/damping settings, etc. The reference signal that is used for deconvolution is typically obtained as the front (or back) surface echo from a suitable sample under conditions identical to those used in acquiring the signal to be processed (deconvolved). When the signal to be processed is acquired from an attenuating medium, the effect of signal attenuation should be appropriately accounted for in the deconvolution technique. If the signal arises from a localized inhomogeneity as in the case of flaw scattered signals, this is easily accomplished by suitably modifying the reference signal; for instance, in the Wiener filter based deconvolution technique [1], the frequency dependent attenuation corresponding to the flaw location is determined and incorporated into the reference signal spectrum. When the inhomogeneities are distributed throughout the material as in the case of grain backscattered signals, the correction for attenuation should vary along the depth of the material. A suitable deconvolution technique for incorporating such correction is based on the Kalman filter [2, 3]. In this technique, the reference signal and the signal to be processed are modeled respectively as the impulse response of a system and the system output. The input to the system is the deconvolved signal that has to be estimated. The Kalman filter algorithm processes the data sequentially and its formulation allows the system parameters to change at each step. This property can be taken advantage of in providing varying amounts of correction for attenuation along the depth of the material.