In this paper we describe the routine photometric calibration of data taken with the VIRCAM instrument on the ESO VISTA telescope. The broadband ZYJHKs data are directly calibrated from 2MASS point sources visible in every VISTA image. We present the empirical transformations between the 2MASS and VISTA, and WFCAM and VISTA, photometric systems for regions of low reddening. We investigate the long-term performance of VISTA+VIRCAM. An investigation of the dependence of the photometric calibration on interstellar reddening leads to these conclusions: (1) For all broadband filters, a linear colour-dependent correction compensates the gross effects of reddening where . (2) For $Z$ and $Y$, there is a significantly larger scatter above E(B-V)=5.0, and insufficient measurements to adequately constrain the relation beyond this value. (3) The $JHKs$ filters can be corrected to a few percent up to E(B-V)=10.0. We analyse spatial systematics over month-long timescales, both inter- and intra-detector and show that these are present only at very low levels in VISTA. We monitor and remove residual detector-to-detector offsets. We compare the calibration of the main pipeline products: pawprints and tiles. We show how variable seeing and transparency affect the final calibration accuracy of VISTA tiles, and discuss a technique, {\it grouting}, for mitigating these effects. Comparison between repeated reference fields is used to demonstrate that the VISTA photometry is precise to better than $\simeq 2\%$ for the $Y$$J$$H$$Ks$ bands and $3\%$ for the $Z$ bands. Finally we present empirically determined offsets to transform VISTA magnitudes into a true Vega system.