Objective. Social interaction theory has received considerable attention. O
ne limitation of previous studies has been that they did not examine how th
e positive effect of racial interaction on white racial tolerance may be en
hanced or reduced by other factors. This article tests five hypotheses deri
ved from the basic assumption of social interaction theory. Methods. Employ
ing EI procedure, a newly developed ecological inference method, this study
uses the precinct-level data from New Orleans' 29 biracial elections to es
timate white crossover voting at the neighborhood level. The hypotheses are
tested by both ordinary least squares and weighted least squares models. R
esults. White crossover voting is positively related to black density. The
positive effect of black density, however, is relatively weak. Other factor
s such as the racial composition of candidate field and election type are m
ore important. Conclusions. Caution needs to be exercised in accepting soci
al interaction theory. The level of racial competition in biracial election
s may reverse white support for black candidates.