This paper describes attempts to determine stabilized emissions of non-road engines without waiting for stable emissions values to be reached, with the goal to shorten laboratory testing time and/or to use real-world, in-service data featuring limited segments of steady-state operating conditions. The emissions from non-road engines are often evaluated and reported in steady-state operating conditions. Many larger engines are tested in the field, due to impracticality of dynamometer testing, resulting in practical limits for testing time at constant operating conditions. With lower fractions of elemental carbon (black soot) in the particulate matter and increased deployment of catalytic aftertreatment devices, longer times are required for reaching stable values. This work seeks to infer stabilized emissions values from limited length segments of unsteady but converging data. Theoretical consideration of thermal factors and storage of material in the exhaust system, and emissions data obtained in a laboratory and on a locomotive engine tested in the field, suggest that the instantaneous emissions in steady-state conditions tend to follow the exponential function y(t) = y(steady) + [y(initial)-y(steady)]*exp(-const.*t) describing both mixing and Newton's law of cooling. Several pathways of non-linear iterative regression has been investigated, generally leading to consistent results, albeit many individual segments of data yield inconsistent or no solution. It appears that when multiple segments of data longer than approximately two minutes are available, there is a high chance at arriving at a plausible steady state value of emissions concentrations. In such case, steady state emissions can be derived from not fully stabilized data, such as from real-world operation of, for example, diesel locomotives, or from large engines tested in a laboratory. With large engines, this can possibly yield considerable savings in testing expenses and improved emissions data. The findings are preliminary, and as of now, interpretation of data requires some skill.