AbstractWhen numerical values of a variable are taken at different points in time, the set of observations constitutes a‘time series’.The variation of tides in a port, the temperature at a given place over a period, the barometric pressure over a series of months, which is a major indication of the weather patterns of an area–these are familiar examples of the kind of phenomena which provide series of values at a succession of points of time.Sets of observations of this kind are usually analysed by means of special techniques described as time series analysis. However, the methods of time series analysis are also applicable and appropriate for the analysis of a variablextwhose values depend on the variablet, whethertrelates to time or linear space.This paper regards the discrepancies between the forward and back runs in a line of geodetic levelling as a‘series of events’or time series. It then uses serial correlation, a technique of time series analysis to examine the behaviour of the discrepancies. The conclusion, which is that serial correlation can explain much of the apparent non-randomness in geodetic levelling observations, sets out to open problems which still remain in the theory of error propagation in levelling and quality control of the operations.