This thesis describes a new tool for automatically seeding faults into Java programs mainly for the purpose of the university education. For each of selected eighty-five types of frequently occurred Java run-time faults, we define fault-seeding principles describing where a specific type of fault could be seeded in a Java program and what possible modifications could be done to generate the fault. Following these pre-defined seeding principles, we build a prototype software system AJFS---Automatic Java Fault Seeder to automatically generate faulty programs with specific number and types of faults seeded. Major technical challenges were successfully solved including building the lexing and parsing module, solving usually ignored characters problem and subsequent three kinds of nondeterminism problems caused by processing usually ignored characters. In this thesis, we also present some samples of implementation results and discuss future extensions.