The analysis of heart rate variability (HRV) aids in the diagnosis of various diseases related to the malfunction of the autonomic nervous system. Traditional approaches for analysis of HRV require the signal to be reasonably stationary during the period of observation. This is not possible when analyzing long duration signals. Detrended fluctuation analysis (DFA) is robust to this issue, as it removes external interferences ("trends") and considers only intrinsic characteristics which are present throughout the signal. DFA is typically performed by segmenting the signal into shorter windows. This has two undesirable effects: (i) if the signal length is not a multiple of the window length, then at least one window will have fewer samples than the others; and (ii) discontinuities are observed on the detrended signal at the edges of each window. Both issues may be addressed using a sliding window. We propose and evaluate this idea, comparing its results with those obtained using the traditional approach. Experiments using different kinds of random and real HRV signals are presented. Statistically significant differences were observed with the proposed approach, especially with respect to α2 values. The proposed method also presented a great reduction in α1 error for white noise, which is a good model for congestive heart failure, with respect to α1 correlations.