AbstractThe fundamental and dominant process operating in all ice‐wedge networks is thermal contraction fracturing. This assumption forms the basis of a numerical model combining fracture initiation and propagation in frozen ground and ice, influence of open fractures on stresses, growth of ice wedges and ground deformation above wedges (Plug and Werner, 2001, 2002). Modelled polygonal networks self‐organise through interactions between fractures, stress and re‐fracture in ice wedges. The resultant polygonal form feeds back on fracturing in individual ice wedges. Spacing, wedge width and fracture frequency in wedges do not reflect mean climate parameters, but instead are sensitive to infrequent climate events and initial conditions, and may vary even under stationary climate — meaning that ice‐wedge casts are difficult to use as estimators of past climate. Burn (2004) suggested that that some of the assumptions underlying the model are incorrect in that they either misrepresent field conditions or ignore crucial site‐specific factors. These criticisms misread and invert the goal of our work, shared in part by any modelling exercise or field investigation, which is to elucidate common, robust behaviours and characteristics across a range of sites rather than to reproduce or describe in precise terms a particular instance. Copyright © 2007 John Wiley & Sons, Ltd.