The so called Luminous Blue Variables (LBVs) form a rather heterogeneous class of very active, luminous (and therefore massive) stars (c.f. R.M. Humphreys, this volume). The following two points are the aim of theoretical work concerning these objects: 1. to know their internal structure from the center up to the atmosphere, i.e. to find out the place of these objects in the course of thermonuclear evolution, or to identify the physical processes which are responsible for the different types of mass loss phenomena, and 2. to conclude which were their preceding evolutionary phases, and to predict their future evolution, i.e. to identify LBV progenitors and descendants, derive their ages, and so on. The basic tool of the theorists is a stellar evolution computer code, which allows to perform evolutionary computations (including different assumptions, which are necessary since several physical ingredients cannot yet be treated from first principles), leading stellar tracks through the part of the HR diagram where LBVs are found. But in this context we meet already a basic problem: in a certain small area of the HR diagram (which may correspond e.g. to some kind of error box) stars of several different types may be identified by the observers, each type may be even devided into several subtypes. Therefore, the correlation of computer models and observed stars is often very difficult or impossible. The various types of stars are mainly classified by spectroscopic criteria, while important “theoretical quantities” as stellar masses or the chemical composition are hard to derive from observations.