In DC plasma spray torches, anode erosion is a common concern. It mainly depends on the heat flux brought by the arc and on the dimensions and residence time of the arc attachment to a given location on the anode wall. The latter depend, to a great extent, on the attachment mode of the arc on the anode wall. This paper compares the anode arc attachment modes predicted by an LTE (Local Thermodynamic Equilibrium) and 2-T (two-temperature) arc models that include the electrodes in the computational domain. It deals with a commercial cascaded-anode plasma torch operated at high current (500 A) and low gas flow rate (60 NLPM of argon). It shows that the LTE model predicted a constricted anode arc attachment that moves on the anode ring, while the 2-T model predicted a diffuse and steady arc attachment. The comparison between the predicted and measured arc voltage showed that the 2-T prediction is closer to the actual voltage. Also, the post-mortem observation of a new anode ring of the actual plasma torch operated under the same conditions for a short time confirmed a diffuse arc attachment on a new anode.