Due to increasing interest in the emissions-producing characteristics of today's automobiles, emissions testing procedures have come under close scrutiny. In addition, development of procedures to measure emissions of vehicles operating in “on-road” conditions have been proposed to gain knowledge of the instantaneous mass flow rates of various legislated gaseous emissions. The problem with the measurement of these instantaneous flow rates is that the responses of modern emissions analyzers to transients are too slow for reliable results. Therefore, a method for improving the dynamic response of these instruments is needed.A method is described which utilizes generalized predictive control theory concepts in conjunction with system identification techniques to produce a software “filter” which reconstructs the distorted output of these analyzers. This “filter” is then applied to actual test data taken from part of an “on-road” emissions and fuel consumption project sponsored by the Federal Highways Administration in conjunction with the Oak Ridge National Laboratory. The procedure for procuring this data involved mapping the operational characteristics of the engine from a test vehicle subjected to “on-road” conditions. These operating points were then simulated on a chassis dynamometer with the post-catalyst exhaust of the vehicle connected to a set of emissions analyzers. Data was taken from these analyzers and passed, off-line, through the reconstruction filter. In addition, part of the Federal Urban Driving Schedule was simulated on the chassis dynamometer and emissions data collected from the test vehicle. For the purpose of this study, only a non-dispersive infra-red (NDIR) analyzer, which is used to measure carbon monoxide (CO), was used.The results that are presented indicate that the generalized predictive control algorithm used in this investigation is valid and reveals a better understanding of the instantaneous concentrations of carbon monoxide emitted from a vehicle operating in real, “on-road” conditions. Furthermore, the results show how the raw, distorted output of the analyzer leads to erroneous assumptions pertaining to the actual production of carbon monoxide, and the specific time in the data record this event occurs.