Spatial release from masking is an important effect in the context of auditory selective attention, since it facilitates, e.g., understanding a target speaker in the presence of a distracting second sound source when target and distractor are spatially separated. Hence, the virtual sound sources created for listening tests using a spatial reproduction method such as higher-order Ambisonics need to be able to successfully recreate this effect. A listening experiment was conducted to test the hypothesis that the spatial release from masking effect is reproduced more accurately with increasing order of Ambisonics reproduction, and to evaluate whether the spatial fidelity of the highest orders tested is sufficient to achieve a degree of realism comparable to real sound sources. In our paradigm, the spatial release from masking effect is quantified as the difference between the measured speech reception thresholds for the co-located and the spatially separated case of target and masking sound sources. Virtual target and masking noise sound sources were created with Ambisonics orders from one to nine using the AllRAD algorithm, reproduced with a spherical loudspeaker array. Real sound sources were included as reference conditions. Implications of the results in the context of an auditory selective attention paradigm are discussed.