Integral models for heavy gas dispersion approximate a dispersing cloud in terms of a small number of variables; each of these is ultimately a function of an independent variable which is usually time (instantaneous releases) or downwind distance (continuous releases). This type of model is used almost exclusively in risk assessment [HSE's risk assessment tool, RISKAT, in: Major Hazards: Onshore and Offshore, October 1992, pp. 607-638; Ann. Rev. Fluid Mech. 21 (1989) 317], but many distinct integral models exist. The code comparison exercise of Mercer et al. [CEA/AEA exchange agreement on external event. Comparison of heavy gas dispersion models for instantaneous releases: final report, Technical Report IR/L/HA/91/6, Health and Safety Laboratory, Sheffield, June 1991; J. Hazard. Mater. 36 (1994) 193] presented the results from a number of integral models in a common format; Mercer found that the range of predictions for some scenarios exceeded three orders of magnitude. Here, the TWODEE shallow layer model [J. Hazard. Mater. 66 (3) (1999) 211; J. Hazard. Mater. 66 (3) (1999) 227; J. Hazard. Mater. 66 (3) (1999) 239] is added to Mercer's code comparison exercise. The physical assumptions used in shallow layer models differ profoundly from those used in integral models and the implications of these differences for risk assessment are discussed. TWODEE was used to simulate four representative cases considered by Mercer. In terms of cloud averaged concentration (CAC) vs. centroid position, the present model gave predictions that were consistent with the integral models used by Mercer. As the model neglects horizontal diffusion for passive clouds, overprediction at large downwind distances was expected, but not generally observed. (C) 2003 Elsevier B.V. All rights reserved.