This paper describes and discusses the malicious agent issues in the agent-mediated electronic markets. We define the malicious agents as software agents that have malicious purpose. Malicious agents can do harm in various ways, such as attacking normal agents, disordering deals, cheating users, stealing intimate information, wasting resources, destroying the markets and so on. We compare the malicious agents with PC viruses in four aspects: originations, actions, means of propagation & infection and speed of propagation & infection. In our opinion, malicious agents are more harmful than the common PC viruses in some measure. We proposed the MA and LMA architecture to deal with the problem described above. Besides, some agent transaction protocols are also presented to show how the MA and LMA architecture works. In the last of this paper, we give some implementation considerations as the foundation for further developments.