The objective of this study is to develop and assess an automatic algorithm based on the peripheral arterial tone (PAT) signal to differentiate between light and deep sleep stages. The PAT signal is a measure of the pulsatile arterial volume changes at the finger tip reflecting sympathetic tone variations and is recorded by an ambulatory unattended device, the Watch-PAT100, which has been shown to be capable of detecting wake, NREM and REM sleep. An algorithm to differentiate light from deep sleep was developed using a training set of 49 patients and was validated using a separate set of 44 patients. In both patient sets, Watch-PAT100 data were recorded simultaneously with polysomnography during a full night sleep study. The algorithm is based on 14 features extracted from two time series of PAT amplitudes and inter-pulse periods (IPP). Those features were then further processed to yield a prediction function that determines the likelihood of detecting a deep sleep stage epoch during NREM sleep periods. Overall sensitivity, specificity and agreement of the automatic algorithm to identify standard 30 s epochs of light and deep sleep stages were 66%, 89%, 82% and 65%, 87%, 80% for the training and validation sets, respectively. Together with the already existing algorithms for REM and wake detection we propose a close to full stage detection method based solely on the PAT and actigraphy signals. The automatic sleep stages detection algorithm could be very useful for unattended ambulatory sleep monitoring assessing sleep stages when EEG recordings are not available.