It is necessary to quantize the variables in VLSI implementation of a turbo decoder. When the quantile interval of received symbol sequence (RSS) is fixed, the required precision of state metrics, log-likelihood ratio, etc are limited by the RSS quantizing error propagation. By the analysis of error propagation, two different quantization schemes are presented for different performance demand without enormous simulations. The error propagation also shows the reason why the excellent results can be obtained with an 8-value look-up table to calculate the correction term when the quantile interval of RSS is 0.25. Also it shows that the max-log-MAP algorithm just like the log-MAP algorithm is the optimal one because of the ignorable correction term for the quantizing error propagation under high signal-noise ratio conditions. The simulation verifies these conclusions.