In this paper, we present a theoretical basis and implementation strategies for 2.5-dimensional (2.5D) sound field reproduction in higher order Ambisonics. The issue associated with 2.5D system is the intrinsic dimensionality mismatch between a desired 2-dimensional (2D) sound field and the loudspeaker employed for reproduction, whose spatial properties are more accurately modelled by that of a 3-dimensional (3D) point source. We formulate this dimensionality mismatch problem in higher order Ambisonics and show that referencing the desired 2D sound field with the corresponding reproduced sound field at the reproduction centre would give the most accurate results. We also derive analytical solutions using 3D point sources as secondary sources for both 2D plane wave and cylindrical wave reproduction. The reproduction accuracy of the proposed method is verified through numerical simulations.