The accreting pulsar GX 301-2 (P = 680 s) has been observed continuous
ly by the large-area detectors of the Burst and Transient Source Exper
iment (BATSE) instrument on the Compton Gamma Ray Observatory since 19
91 April 5. Orbital parameters determined from these data are consiste
nt with previous measurements, with improved accuracy in the current o
rbital epoch. The most striking features in the pulsar frequency histo
ry are two steady and rapid spin-up episodes, with (v) over dot approx
imate to (3-5) x 10(-12) Hz s(-1) each lasting for about 30 days. They
probably represent the formation of transient accretion disks in this
wind-fed pulsar. Except for these spin-up episodes, there are virtual
ly no net changes in the neutron star spin frequency on long timescale
s. We suggest that the long-term spin-up trend observed since 1984 ((v
) over dot approximate to 2 x 10(-13) Hz s(-1)) may be due entirely to
brief (approximate to 20 days) spin-up episodes similar to those we h
ave discovered. We assess different accretion models and their ability
to explain the orbital phase dependence of the observed flux. In addi
tion to the previously observed preperiastron peak at orbital phase 0.
956 +/- 0.022, we also find a smaller peak close to apastron at orbita
l phase 0.498 +/- 0.057. We show that if the companion star's effectiv
e temperature is less than 22,000 K, then it must have a mass M-c < 70
M. and a radius R-c < 85 R. so as not to overfill the tidal lobe at p
eriastron. In order not to overflow the Roche lobe at periastron, the
corresponding values are M-c < 55 M. and R-c < 68 R.. These constraint
s are nearly at odds with the reclassification by Kaper et al. of the
companion as a B1 Ia+ hypergiant.