We investigate the clustering of galaxies selected in the 3.6 micron band of the Spitzer Wide-area Infrared Extragalactic (SWIRE) legacy survey. The angular two-point correlation function is calculated for eleven samples with flux limits of S_3.6 > 4
-400 mujy, over an 8 square degree field. The angular clustering strength is measured at >5-sigma significance at all flux limits, with amplitudes of A=(0.49-29)times10^{-3} at one degree, for a power-law model, Atheta^{-0.8}. We estimate the redshift distributions of the samples using phenomological models, simulations and photometric redshifts, and so derive the spatial correlation lengths. We compare our results with the GalICS (Galaxies In Cosmological Simulations) models of galaxy evolution and with parameterized models of clustering evolution. The GalICS simulations are consistent with our angular correlation functions, but fail to match the spatial clustering inferred from the phenomological models or the photometric redshifts. We find that the uncertainties in the redshift distributions of our samples dominate the statistical errors in our estimates of the spatial clustering. At low redshifts (median z<0.5) the comoving correlation length is approximately constant, r_0=6.1pm0.5h^{-1} Mpc, and then decreases with increasing redshift to a value of 2.9pm0.3h^{-1} Mpc for the faintest sample, for which the median redshift is z=1. We suggest that this trend can be attributed to a decrease in the average galaxy and halo mass in the fainter flux-limited samples, corresponding to changes in the relative numbers of early- and late-type galaxies. However, we cannot rule out strong evolution of the correlation length over 0.5<z<1.
