Acoustic measures of vocal function are attractive due to their nonintrusive nature and due to the ease with which they can be obtained. In this paper, we developed an acoustic measure based on linear prediction modeling and filterbank analysis of continuous speech samples. The input speech sample was first modeled using a combination of short-term and long-term all pole linear prediction filters. The input speech sample and the residual signal were then divided into subbands using cosine-modulated or gammatone filterbanks. The Signal-to-Residue Ratio (SRR) was calculated as the weighted combination of the ratios of the input and residual signal energies in the subbands. The performance of the SRR parameter was evaluated with speech samples collected from patients suffering from vocal fold cancer before and after radiation therapy. Results showed that the SRR measure correlates better with the perceptual judgments of voice quality than the global SNR parameter.