I study the sensitivity of three-dimensional Eulerian simulations of the evolution of 26 trace chemical species. The region simulated is 2,400 by 2,400 km horizontally and from the surface to about 15 km vertically. The uncertainty in simulated concentration is from perturbations or errors in input parameters and physical parameterizations: chemistry, emissions, meteorology, and initial and boundary conditions. I vary each input parameter p and calculate the root-mean-square (rms) change in concentration &sgr; between the sensitivity run and the base run. The sensitivity parameter, S=∂&sgr;/∂p, is defined for each species and input parameter. For three scenarios I assign errors to each input parameter. For each species the total error in simulated concentration, &sgr;t, is the rms sum of the contributions by errors in each input parameter. I describe this error as the uncertainty, un=&sgr;t/cb, where cb is the horizontal mean concentration in the base run, and as the normalized uncertainty, nu=&sgr;t/&sgr;b, where &sgr;b is the standard deviation of fluctuations horizontally about cb. The uncertainty suggests that O3 is well simulated, but the normalized uncertainty shows it is poorly simulated. I prefer nu because it is more nearly constant for different species, but nu increases with height, implying that the simulations are poorer aloft. If the input parameters have 25% errors, then, averaged for all species, un=0.53¿0.43 and nu=0.45¿0.24. Often, &sgr;t is largely from errors in meteorology and in initial and boundary conditions. Chemical errors dominate steady state species. Initial and boundary conditions must have errors less than 25% before the errors in emissions begin to be important. ¿ American Geophysical Union 1988