Sleep is an important topic in the lives of human beings because it is necessary for survival. Other than its general known function of revitalizing the human body sleep can also be associated with various pathologic conditions. Therefore sleep has been subject of studies for many of years. It can be analyzed using Polysomnography (PSG). In PSG multiple instruments (including electroencephalography, electromyography, electrooculography and electrocardiography (ECG)) are required to record various biosignals in order to identify and analyze different sleep phases and sleep stages. Additionally the recorded signals have to be assessed visually by an expert. It is desirable to reduce the amount of instrumentation required for the analysis of sleep quality. During sleep the autonomous nervous system (ANS) regulates various bodily functions. Besides for example respiration, digestion, vasomotor activity and reflex actions the ANS also effects cardiac regulation. There is close relation between autonomic control of heart rate and the central nervous system, especially during sleep when ambient factors do not dominate. The variation in time intervals of consecutive heart beats is called heart rate variability (HRV) and reflects the activity of the ANS. Therefore it might be possible to analyze sleep quality using HRV. Our objective was to assess a new method to identify sleep stages by using HRV in order to evaluate sleep without the use of multiple devices required by traditional PSG. There are various linear and nonlinear methods to analyze HRV. Each method has its advantages, disadvantages and limitations. We used spectrum weighted mean frequencies of the total HRV spectrum of the power spectral density (PSD) to identify different sleep stages and assessed how they correlate to somnograms recorded by multiple devices of PSG. The correlation was analyzed by calculating corresponding cross correlation coefficients. For our analysis we acquired 22 datasets of raw ECG signals from a sleep laboratory and their somnograms recorded by conventional PSG. Because of baseline drift and noise, preprocessing of the ECG signals was necessary in order to ensure a good R-peak detection. Because of their effect on PSD, outliers originating from ectopic beats and falsely identified R-peaks had to be corrected. We used different options (Removal of sleep stages of short periods of time, combination of sleep stages, different inversion methods and different filter options) to analyze and improve calculated cross correlation coefficients. Our results showed that some somnograms, especially those from healthy subjects, had a high correlation with our HRV based sleep estimation. Semi-periodic somnograms provide better results in terms of the HRV-based sleep prediction than fragmented somnograms. The combination of sleep stages (Wake + Rapid eye Movement, S1 + S2, S3 + S4) seems to improve the HRV-based estimation of somnograms, whereas the combination of only S3+S4 seems to weaken this estimation in most cases. The removal of sleep stages with the duration of only a few minutes does not seem to alter our results significantly. We found that the negation of the weighted mean frequency tends to yield better sleep prediction than its inversion. The preprocessing methods we used did not significantly influence our results. We conclude that HRV can be used to assess sleep quality in many cases, especially in rather healthy somnograms. However, further research is needed to improve sleep prediction in fragmented somnograms.