The education of researchers in Sweden is regulated by a nationwide reform implemented in 1969, which intended to limit doctoral programs to 4 years without diminishing quality. In an audit performed by the government in 1996, however, it was concluded that the reform had failed. Some 80% of the doctoral students admitted had dropped out, and only 1% finished their PhD degree within the stipulated 4 years. In an attempt to determine the causes of this situation, we singled out a social-science department at a major Swedish university and interviewed those doctoral students who had dropped out of the program. This department was found to be representative of the nationwide figures found in the audit. The students interviewed had all completed at least 50% of their PhD studies and had declared themselves as dropouts from this department. We conclude that the entire research education was characterized by a laissez-faire attitude where supervisors were nominated but abdicated. To correct this situation, we suggest that a learning alliance should be established between the supervisor and the student. At the core of the learning alliance is the notion of mutually forming a platform form which work can emerge in common collaboration. The learning alliance implies a contract for work, stating its goals, the tasks to reach these goals, and the interpersonal bonding needed to give force and endurance to the endeavor. Constant scrutiny of this contract and a mutual concern for the learning alliance alone can contribute to its strength.