This paper discusses the test problem
concerning ratios of two normal density functions. We present large
sample tests based on asymptotic normality of some statistics and
the test based on generalized $p$-values, respectively. Some
frequency properties of the generalized $p$-values are proved. We
conclude from simulation results that the performance of the test
based on generalized $p$-values is better than the large sample
tests in small sample cases.