Cerebral blood flow can be measured in neonates by near infra-red spectrophotometry. The tracer is oxyhaemoglobin. The purpose of the study is to compare the test-retest variability of two previously proposed methods (UCH and COP) of analysis, and to investigate the influence of sampling rates, smoothing and integration periods. Under clinical conditions good measurements are often difficult to obtain. Therefore, a second goal is to find ways of determining the quality of individual measurements. 380 cerebral blood flow measurements from 69 infants are analysed. The data set is optimised statistically for the lowest test-retest variability and the following results are obtained. The test-retest variability of measurements at 2 s sampling time data is considerably worse than at 0.5 s sampling time. Smoothing does not change the test retest variability. A 6 s integration period gives higher values and higher test-retest variability than an 8 s integration period. By applying the suggested criteria, a test-retest variability of 17% is achieved, if 50% of the measurements are rejected. The mean cerebral blood flow is 12.2 ml (100 g)-1 min-1 for the UCH method and 9.7 ml [corrected] (100 g)-1 min-1 for the COP method. The test-retest variability of both methods is comparable for 0.5 s sampling time. For 2 s sampling time the method proposed by Skov et al. is significantly better. These test retest variabilities represent maximum values, part of the observed variability may be due to physiological changes of unknown magnitude.