Head-and-neck tumors are often situated at an air-tissue interface what may result in an underdosage of part of the tumor in radiotherapy treatments using megavoltage photons, especially for small fields. In addition to effects of transient electronic disequilibrium, for these small fields, an increased lateral electron range in air will result in an important extra reduction of the central axis dose beyond the cavity. Therefore dose calculation algorithms need to model electron transport accurately. We simulated the trachea by a 2 cm diameter cylindrical air cavity with the rim situated 2 cm beneath the phantom surface. A 6 MV photon beam from an Elekta linear accelerator, equipped with the standard multileaf collimator (MLC), was assessed. A and a field, both widthwise collimated by the MLC, were applied with their long side parallel to the cylinder axis. Central axis dose rebuild-up was studied. Radiochromic film measurements were performed in an in-house manufactured polystyrene phantom with the films oriented either along or perpendicular to the beam axis. Monte Carlo simulations were performed with BEAM and EGSnrc. Calculations were also performed using the pencil beam (PB) algorithm and the collapsed cone convolution (CCC) algorithm of Helax–TMS (MDS Nordion, Kanata, Canada) version 6.0.2 and using the CCC algorithm of Pinnacle (ADAC Laboratories, Milpitas, CA, USA) version 4.2. A very good agreement between the film measurements and the Monte Carlo simulations was found. The CCC algorithms were not able to predict the interface dose accurately when lateral electronic disequilibrium occurs, but were shown to be a considerable improvement compared to the PB algorithm. The CCC algorithms overestimate the dose in the rebuild-up region. The interface dose was overestimated by a maximum of 31% or 54%, depending on the implementation of the CCC algorithm. At a depth of 1 mm, the maximum dose overestimation was 14% or 24%.