We describe the construction of MegaZ-LRG, a photometric redshiftcatalogue of over one million luminous red galaxies (LRGs) in theredshift range 0.4 < z < 0.7 with limiting magnitude i < 20.The catalogue is selected from the imaging data of the Sloan Digital SkySurvey (SDSS) Data Release 4. The 2dF-SDSS LRG and Quasar (2SLAQ)spectroscopic redshift catalogue of 13000 intermediate-redshift LRGsprovides a photometric redshift training set, allowing use of ANNz, aneural network-based photometric-redshift estimator. The rms photometricredshift accuracy obtained for an evaluation set selected from the 2SLAQsample is σz = 0.049 averaged over all galaxies, andσz = 0.040 for a brighter subsample (i < 19.0). Thecatalogue is expected to contain ~5 per cent stellar contamination. TheANNz code is used to compute a refined star/galaxy probability based ona range of photometric parameters; this allows the contaminationfraction to be reduced to 2 per cent with negligible loss of genuinegalaxies. The MegaZ-LRG catalogue is publicly available on the WorldWide Web from http://www.2slaq.info.