The pulsar PSR B1259–63 is in a highly eccentric 3.4-yr orbit with the Be star SS 2883. Timing observations of this pulsar, made over a 7-yr period using the Parkes 64-m radio telescope, cover two periastron passages, in 1990 August and 1994 January. The timing data cannot be fitted by the normal pulsar and Keplerian binary parameters. A timing solution including a (non-precessing) Keplerian orbit and timing noise (represented as a polynomial of fifth order in time) provides a satisfactory fit to the data. However, because the Be star probably has a significant quadrupole moment, we prefer to interpret the data by a combination of timing noise, dominated by a cubic phase term, and ω. and x. terms. We show that the ω. and x. terms are likely to be a result of a precessing orbit caused by the quadrupole moment of the tilted companion star. We further rule out a number of possible physical effects which could contribute to the timing data of PSR B1259–63 on a measurable level.