diff --git a/doc/texinfo/gmsh.texi b/doc/texinfo/gmsh.texi
index 33236188fcec642e32712e93888392d7a929d0bd..229d5a6bca6e199c2a2095aa6b1e6e65de8277c1 100644
--- a/doc/texinfo/gmsh.texi
+++ b/doc/texinfo/gmsh.texi
@@ -1,5 +1,5 @@
 \input texinfo.tex @c -*-texinfo-*-
-@c $Id: gmsh.texi,v 1.72 2003-07-07 16:43:46 geuzaine Exp $
+@c $Id: gmsh.texi,v 1.73 2003-07-19 23:35:44 geuzaine Exp $
 @c
 @c Copyright (C) 1997-2003 C. Geuzaine, J.-F. Remacle
 @c
@@ -491,8 +491,8 @@ commands and command arguments to depend on previous calculations (see
 @item
 generate 1D, 2D and 3D simplicial (i.e.@: using line segments, triangles and
 tetrahedra) finite element meshes. The performance of the 1D and 2D
-algorithms is pretty good; the 3D algorithm is still somewhat experimental
-and slow (see @ref{Mesh module}, and @ref{Tutorial});
+algorithms is pretty good; the 3D algorithm is still experimental and slow
+(see @ref{Mesh module}, and @ref{Tutorial});
 @item
 specify target element sizes accurately. Gmsh provides several mechanisms to
 control the size of the elements in the final mesh: through interpolation
@@ -561,6 +561,11 @@ conforming in the sense of finite element meshes;
 the user has no control over the quality of the mesh elements generated by
 the 3D unstructured algorithm;
 @item
+the 2D anisotropic and the 3D unstructured algorithms are still experimental
+and not very robust.  If these algorithms fail, try to change some
+characteristic lengths to generate meshes that better suit the geometrical
+details of the structures;
+@item
 Gmsh was designed to solve academic test-cases, not industrial-size
 problems. You may find that Gmsh is too slow for large problems (with
 thousands of geometric primitives, or millions of mesh/post-processing