This paper presents how to control the input bit-width of a turbo decoder according to the signal-to-noise ratio (SNR) adaptively. It is crucial to minimize the input bit-width while maintaining the error-correcting performance. Several quantization schemes have been presented to properly decide the input resolution of turbo decoding, but they all assume a fixed bit-width irrespective of the SNR. This paper proposes a new method to adaptively change the quantization bit-width with respect to the SNR. In addition, this paper suggests a novel method to estimate the frame-error rate (FER) performance of a turbo decoder dealing with quantized channel outputs. As a result, the whole SNR range is divided into a number of regions each of which can be processed with a certain quantization bit-width.