The paper presents the development and validation of phenomenological predictive schemes for quasi-dimensional modeling of pollutant emissions in direct injected Diesel engines. Models for nitric oxide (NO), carbon monoxide (CO), as well as soot and unburned hydrocarbons (HC) have been developed. All of them have been implemented into a DI Diesel engine simulation environment, previously developed by the authors, which features quasi-dimensional modeling of spray injection and evolution, air-fuel mixture formation, as well as auto-ignition and combustion. An extended Zel'dovich mechanism, which takes into account the three main, thermal-NO formation chemical reactions has been developed for predicting NO emissions. A simple, one-reaction soot formation model has been implemented, while a new approach has been proposed for soot oxidation, which considers two different temperature ranges: the well-established Nagle and Strickland-Constable one has been adopted for the highest temperatures, while a new, single-step reaction model has been implemented at the low temperatures. The model for carbon monoxide formation relies on five chemical reactions, whose kinetics are computed exploiting partial equilibrium assumptions, in a system of 11 species. Finally, hydrocarbon emissions have been modeled taking into account the effects of three main sources: fuel injected and mixed beyond the lean combustion limit, fuel yielded by the injector sac volume and nozzle holes, as well as overpenetrated fuel. A detailed comparison with experimental data from a high speed, direct injected diesel engine, carried on for both full and partial load and for a wide range of engine speeds, shows that the models are capable to predict the engine emissions with reasonable reliability.