There is growing interest in the concept of learning design (IMS-D 2003, Britain 2004). However there is an unfortunate tendency to treat learning objects as "things" slotted into learning designs, rather than being built on learning designs themselves. Britain points out that, in order to achieve the learning objective, a learning object has to have a learning design built in (however basic). The key concept of generative learning objects is to separate the deep structure from the surface structure of the learning object. In that sense it uses a model of reuse, which is closer to that in object-oriented software engineering. This raises the question of what exactly is the nature of the deep structure. As argued in Boyle et al. (2004), this includes the learning design that underpins the object. The concept of GLOs separates the learning design from its surface instantiation. The surface learning object is then viewed as a particular realisation of the underlying learning design. This has a number of advantages. Importantly it focuses attention on the quality of the learning design underpinning the object. As the surface object is generated in a series of steps from this design, it permits intervention at these stages. This means that many variants can be produced from the same design. Because the learning objects are divided into their basic components, and structures, it becomes easier to identify and modify individual components.