A reduced order kinetic model for NO (Nitric Oxide) prediction, based on the virtual chemistry methodology [1], is developed and applied. Virtual chemistry aims to optimize thermochemical properties and kinetic rate parameters of a network of virtual species and reactions. A virtual main chemical mechanism is dedicated to temperature and heat release prediction and is coupled with the flow governing equations, whereas satellite sub-mechanisms are designed to predict pollutants formation. Two virtual chemistry mechanisms are here employed: a main mechanism for calculating the temperature and heat release rate and a second mechanism dedicated to NO prediction. To recover the chemical structure of multi-mode combustion, both premixed and non-premixed flamelets are included in the learning database used to optimize the virtual NO mechanism. A multi-zone optimization procedure is developed to accurately capture both fast and slow NO chemistry that include prompt, thermal and reburning pathways. The proposed NO sub-mechanism and optimization methodology are applied to CH4/air combustion. Laminar 1-D premixed and non-premixed flamelet configurations are first tested. The approach is then further assessed in 2-D CFD laminar flame simulations, by providing a direct comparison against detailed chemistry. 2-D premixed, non-premixed and partially premixed flame configurations are numerically investigated. For all cases, the virtual mechanism fairly captures temperature and NOx chemistry with only 12 virtual species and 8 virtual reactions with a drastic CPU time reduction compared to detailed chemistry.