This paper is concerned with the problem of load-dependent H2 control for vehicle semi-active suspension. A quarter-car model equipped with a magnetorheological (MR)-damper, which captures essential features of a real car suspension, is considered in this study. H2-norm measures the root-mean-square (RMS) value of output to white noise input. Considering the fact that road roughness is often modelled as white noise, H2-norm is used to quantify control objectives of ride comfort and safety as well as suspension deflection and control effort. To guarantee system performance against parameter variations, a linear matrix inequality (LMI)-based design framework has been utilised and a linear parameter-varying (LPV) controller is synthesised. The design procedure of the semi-active suspension requires the inverse dynamics of MR damper, which is obtained through a locally linear neuro-fuzzy (LLNF) network. To illustrate the effectiveness of the proposed approach, the system outputs for both impulse and real road inputs are compared with H∞ controller in terms of performances.