Object-oriented development is expected to provide many benefits, but observations of industrial practice show that there are many problems to overcome. This paper discusses two issues: (1) The architectural issue of how to connect database, application, and screen objects. (2) To what extent expected benefits such as seamless transition from analysis to design, improved usability, etc. can be obtained. The architectural issue appears to have a significant influence on the expected benefits. The paper is based on studies of about seven experienced, but very different, development teams, plus a lot of casual observations. I will present the three basic architectures that I found in practice and discuss their advantages and problems. Surprisingly, only one team had a solid architecture that solved most of the problems.