IntrodnctionI n the 1970's, Nonsmooth Analysis provided an effective approach to a large class of optimization problems involving functions that can be approximated by convex sets of linear continuous maps having some specific properties: generalized gradient and generalized Jacoman matrix of CLARKE [8, 9, 10, 111, derivate container of WARGA [52], screen, shield and fan of HALKIW [lS, 191, generalized derivative of POURCIAU [46], shield of SWEETSER [51]. Later, i t was proven by IOFFE [ 2 5 , 26, 291 that many stronger results concerning the differential calculus for single-valued maps can be obtained if set-valued maps rather than conves sets of linear continuous maps are used as local approximations. Also, multivalued differential was shown to be very useful for stability theory of differential equations [30]. On the other hand, without the help of set-valued maps many important problems arising in control theory, mathematical economics, stability theory of differential inclusions and some other fields of applied mathematics could not be modelled and studied. It is then natural to ask whether one can introduce concepts of derivatives of set-valued maps t o allow the correspondmg theories of differentiability to be applied to a broad class of problems involving set-valued maps. In answer to this question two approaches have been suggested. The first approach proposed in [4,7,13,14,15,24,32,34,50]deals withlocalapproxjmations of a set-valued map T a t a point xo belonging to the domain of T. The class of maps that are differentiable in the sense given in almost all these papers seems to be restrictive and many typical theorems of classical calculus (such as surjectivity theorems, implicit function theorems) w h c h play a crucial role e.g. in optimization theory have not been generalized to set-valued maps. The main idea of the second approach followed by METHLOUTHI [33], ATTBIN [ 1, 2, 31, OETTLI [36] and SACH [35] is that T is identified with its graph (denoted by gr 2') a i d any cleri~atii-e of T is defined at a point zoEgr T as a map whose graph Q is tangent to gr T a t zo in snme sense. In [33] Q is the graph of a linear continuous map