Crop models can be evaluated based on accuracy in simulating several S
ears' yields for one location or on accuracy in simulating long-term m
ean yields for several locations. Our objective was to see how the ALM
ANAC (Agricultural Land Management Alternatives with Numerical Assessm
ent Criteria) model and a new version of CERES-Maize (Crop Environment
Resource Synthesis) simulate grain yield of rainfed maize (Zea mays L
.). We tested the models at one county in each of nine states: Minneso
ta, New York, Iowa, Illinois, Nebraska, Missouri, Kansas, Louisiana, a
nd Texas (MN, NY, IA, IL, NE, MO, KS, LA, and TX). Simulated gain yiel
ds were compared with grain yields reported by the National Agricultur
al Statistical Service (NASS) for 1983 to 1992. In each county we thos
e a soil commonly used in maize production, and we used measured weath
er data. Mean simulated grain yield for each county was always within
5% of the mean measured gain yield for the location. Within locations,
measured grain yield was regressed on simulated grain yields and test
ed to see if the slope was significantly different from 1.0 and if the
y-intercept was significantly different from 0.0, both at the 95% con
fidence level. Only at MN, NY, and NE for ALMANAC and at MN, NY, and T
X for CERES was slope significantly different from 1.0 or intercept si
gnificantly different from 0.0. The CVs of simulated grain yields were
similar to the those of measured yields at most sites. Also, both mod
els were appropriate for predicting an individual year's yield for mos
t counties. Values for plant parameters, such as heat units for develo
pment and the harvest index, and values for soil parameters describing
soil water-holding rapacity offer users reasonable inputs for simulat
ing maize grain yield over a wide range of locations.