A two-scale analytical-numerical homogenisation approach is developed to predict effective elastic properties of ultra high performance fibre reinforced concrete considering distribution of pore sizes acquired from 3D micro X-ray computed tomography (μXCT) images of 24.8μm resolution. In the first scale, the mortar, consisting of sand, cement paste and a large number of small pores (10-600μm), is homogenised using analytical Mori-Tanaka method with constituents' moduli from micro-indentation. In the second, μXCT images of a 20mm cube are converted to mesoscale representative volume elements for finite element homogenisation, with fibres and a small number of large pores (≥600μm) in the homogenised mortar. The resultant elastic moduli are compared favourably with experimental data. This approach accounts for a large number of pores with a wide size range yet without excessive computational cost. Effects of fibre volume fraction and orientation are investigated, demonstrating the approach's potential to optimise the material's micro-structure for desired properties.