We present a method for the identification of heavily absorbed active galactic nuclei (AGN) (NH > 1023 cm−2) from X-ray photometric data. We do this using a set of XMM–Newton reference spectra of local galaxies for which we have accurate NH information, as described in Brightman & Nandra. The technique uses two rest-frame hardness ratios which are optimized for this task, which we designate HR1 (2–4/1–2 keV) and HR2 (4–16/2–4 keV). The selection method exploits the fact that while obscured AGN appear hard in HR2 due to absorption of the intrinsic source flux below ∼4 keV, they appear soft in HR1 due to excess emission originating from scattered source light, thermal emission or host galaxy emission. Such emission is ubiquitous in low-redshift samples. The technique offers a very simple and straightforward way of estimating the fraction of obscured AGN in samples with relatively low signal-to-noise ratio in the X-ray band. We apply this technique to a moderate redshift (z∼ 1) sample of AGN from the Chandra Deep Field-North, finding that 61 per cent of this sample has NH > 1023 cm−2. A clear and robust conclusion from our analysis is that in deep surveys the vast majority of sources do not show hardness ratios consistent with a simple absorbed power law. The ubiquity of complex spectra in turn shows that simple hardness ratio analysis will not yield reliable obscuration estimates, justifying the more complex colour–colour analysis described in this paper. While this method does very well at separating sources with NH > 1023 cm−2 from sources with lower NH, only X-ray spectroscopy can identify Compton-thick sources, through the detection of the Fe Kα line. This will be made possible with the high throughput X-ray spectral capabilities of Athena.