It depends on the element formulation. If there's uncoupling amongst different DOFs it is normal to have zero valued elements in the element stiffness matrix outside the diagonal.

 

The example is a simple 1D truss which is fixed at one end and subjected to a force at the other end. So the stiffness matrix, I think, should be a banded matrix.
In this example, the nodes are number from left to right as 1,2,3,4,5,6. But the stiffness shows that A22=A33=A55=A66=0 and others such as A12=A21=A23=A32=...=A65=0. I thinks there terms should not be zero.

Thanks.

The manual is right. Three degrees of freedom areactivated and so for two-noded truss element, the stiffness matrix is a 6x6 matrix.

Thank you very much mapi. I had been looking around like crazy to find how you define your own stiffness matrix in abaqus and your pointer to the uellinea.inp got me on the right way. It soo simple and obvious when you see it but I couldnt find it in the manual. I always ended up in substructure where you can print out the stiffness matrix but not write in your own.
I write the syntax here so other might find it also if they search.
(from the file uellinea.inp under 4.1.11 in verification manual)
*USER ELEMENT,NODES=2,TYPE=U1001,LINEAR,UNSYMM
1,2,3
*MATRIX,TYPE=STIFFNESS
 1.05E9, 0.    , 0.    ,-1.05E9
 0.    , 0.
**
 0.    , 0.    , 0.    , 0.
 0.    , 0.
**
 0.    ,       , 0.0   , 0.
 0.    , 0.
**
-1.05E9, 0.    , 0.    , 1.05E9
 0.    , 0.
**
 0.    , 0.    , 0.    , 0.
 0.    , 0.
**
 0.    , 0.    , 0.    , 0.
 0.    , 0.0