Deterministic models are used to determine the effect of quantization errors in radix-2 DIT-FHT and DIF-FHT algorithms. The results are then supplemented assuming a particular statistical model. Overall scaling and step-by-step scaling are treated independently for both DIT and DIF-FHT. The problem of twiddle factor quantization is also considered independently in both DIT and DIF-FHT algorithms. The expected average error power is derived for all the above-mentioned cases and comparisons are made with the FFT for each case. © 2004 Elsevier B.V. All rights reserved.