Air pollution associated with road transport is a major environmental issue in urban areas. Buildings in urban areas are the artificial obstacles to atmospheric flow and cause reduced ventilation for street canyons. For a deep street canyon, there is evidence of the formation of multiple segregated vortices, which generate flow regimes such that pollutants exhibit a significant contrast between these vortices. This results in poor air ventilation conditions at pedestrian level, thereby leading to elevated pollutant levels and potential breaches of air quality limits. The hypothesis of a well-mixed deep street canyon in the practical one-box model approach is shown to be inappropriate. This study implements a simplified simulation of the canyon volume: a coupled two-box model with a reduced chemical scheme to represent the key photochemical processes with timescales similar to and smaller than the turbulent mixing timescale. The two-box model captures the significant pollutant contrast between the lower and upper parts of a deep street canyon, particularly for NO2. Core important parameters (i.e. heterogeneity coefficient, exchange velocity and box height ratio) in the two-box model approach were investigated through sensitivity tests. The two-box model results identify the emission regimes and the meteorological conditions under which NO2 in the lower canyon (i.e. the region of interest for the assessment of human health effects) is in breach of air quality standards. Higher NO2 levels were observed for the cases with higher heterogeneity coefficients (the two boxes are more segregated), with lower exchange velocities (worse ventilation conditions), or with smaller box height ratios (reduced dilution possibly due to secondary smaller eddies in the lower canyon). The performance of a one-box model using the same chemical scheme is also evaluated against the two-box model. The one-box model was found to systematically underestimate NO2 levels compared with those in the lower box of the two-box model for all test scenarios. This underestimation generally tends to worsen for higher heterogeneity coefficients, lower exchange velocities or smaller box height ratios. This study highlights the limitation of the assumption of homogeneity in single box models for street canyon simulation, and the inherent uncertainties that must be borne in mind to appropriately interpret such model output (in particular, that a single-box treatment will systematically underestimate NO2 as experienced at street level).