In this paper we report on the results of the Blind Test 2 workshop, organized by Norcowe and Nowitech in Trondheim, Norway in October 2012. This workshop was arranged in order to find out how well wind turbine simulation models perform when applied to two turbines operating in line. Modelers with a suitable code were given boundary conditions of a wind tunnel test performed in the large wind tunnel facility at the Department of Energy and Process Engineering, at NTNU Trondheim, where two almost identical model turbines with a diameter of about 0.9 similar to m had been tested under various operating conditions. A detailed geometry specification of the models could be downloaded and the modelers were invited to submit the calculation without knowing the experimental results in advance. Nine different contributions from eight institutions were received, representing a wide range of simulation models, such as a LES coupled with an actuator line rotor model, RANS using an actuator disc, U-RANS models applied to fully resolved turbine model geometries, as well as a vortex panel method. The comparison showed a larger than expected scatter on the performance calculation of the upstream turbine (+/- 20%), and an even higher uncertainty for the downstream turbine, especially at operating conditions close to the runaway point. The modelers were requested to document the wake development downstream of the second turbine, the development behind the first turbine had been the challenge for a previous blind test (see Krogstad and Eriksen [17]). Mean flow calculations reported at X = 1D downstream of the second turbine showed that the models which fully resolved boundary layers on the rotor surface performed best. Including the tower and the hub in the simulation improved the accuracy of the predictions and is essential in capturing the important asymmetries that develop in the wake. These turbine details strongly influence the development near the center of the wake, but are often omitted in simulations in order to incorporate simplifying symmetry conditions in the calculations. Further from the rotor, at X = 4D, the LES simulations coupled to actuator line rotor models performed well and were able to capture the main features of the mean and turbulent flows, while RANS models using actuator disc models showed limitations especially in predicting correctly the turbulent kinetic energy. (C) 2014 Elsevier Ltd. All rights reserved.