A Support Vector Machine (SVM in short) is a mechanism
for data analysis such as
classification, regression, and clustering.
In this paper,
we consider the soft margin formulation of SVM
for two-class classification problems,
and discuss a way to choose an appropriate parameter
that determines the influence of misclassified examples.
Based on the observation in \cite{bennett-bredensteiner:icml00},
we propose one criterion for choosing the parameter,
and give some fact justifying this criterion.