diff --git a/doc/salome/gui/GEOM/images/tolerances.png b/doc/salome/gui/GEOM/images/tolerances.png
new file mode 100644
index 000000000..69b136767
Binary files /dev/null and b/doc/salome/gui/GEOM/images/tolerances.png differ
diff --git a/doc/salome/gui/GEOM/input/faq.doc b/doc/salome/gui/GEOM/input/faq.doc
index 78fa0d58c..3976f44ea 100644
--- a/doc/salome/gui/GEOM/input/faq.doc
+++ b/doc/salome/gui/GEOM/input/faq.doc
@@ -6,6 +6,7 @@ Here you can find the answers to some frequently asked questions:
 
 <ul>
 <li>\subpage partition_explanation "What is the difference between partition, compounds and fuse operation ?" </li>
+<li>\subpage size_models_range "What are the valid sizes of models ?" </li>
 </ul>
 
 
diff --git a/doc/salome/gui/GEOM/input/size_models_range.doc b/doc/salome/gui/GEOM/input/size_models_range.doc
new file mode 100644
index 000000000..5f77a22e3
--- /dev/null
+++ b/doc/salome/gui/GEOM/input/size_models_range.doc
@@ -0,0 +1,87 @@
+/*!
+
+\page size_models_range Sizes of Models in Salome
+
+\tableofcontents
+
+This document determines the range of numbers (tolerances, locations
+and sizes) that are to be taken into account for any 3D model design
+in Salome. Although it is not obligatory to create models within this range,
+algorithms can fail or return unexpected result in this case.
+
+This document refers mainly to Open CASCADE Technology (OCCT). However it
+concerns Salome as well as OCCT is a modeling core of Salome %GEOM module.
+
+Any model in 3D-space has its location and sizes. The last two things in Salome
+and OCCT are represented by the double precision floating point numbers.
+
+The goal of the document is to define the range of numbers that can be used in
+modeling algorithms provided by Salome and Open CASCADE Technology.
+
+\section sec1 Maximal Size of the Model
+
+The Maximal Size of the model is a number defined as the maximal diameter of
+enclosed sphere built for the model. In OCCT any model has a location defined
+relative the absolute origin. Thus the maximal diameter above should be built
+taking into account the model itself and its location.
+
+In Open CASCADE there are two tolerances: Tolerance Confusion (TolC)
+and Tolerance Angular (TolA) (see OCCT Precision package for more details).
+These values are used for geometric comparisons. They are not used inside
+low-level algorithms (intersection for e.g.), where more precise values are
+used instead. The value TolC guarantees that the error associated with
+the computations for given geometric entity is not greater than TolC.
+
+- TolC - precision value when checking coincidence of two points
+  [by default 1.e-7];
+- TolA - precision value when checking the equality of two angles
+  [by default 1.e-12].
+
+For more information on tolerance definition please see
+<a href="SALOME_BOA_PA.pdf">Chapter 4 of this document</a>. To see limitations
+that are due to modeling errors or inaccuracies of tolerance usage please
+refer to <a href="SALOME_BOA_PA.pdf">Chapter 9.2.2 of the same document</a>.
+
+To provide robust geometric modeling the computations should be consistent,
+i.e. the one tolerance value should be used for all computations. To provide
+consistent computations the values TolC and TolA should be consistent:
+
+<CENTER><B><PRE>Smax = TolC / TolA             (1)</PRE></B></CENTER>
+
+\image html tolerances.png "TolC, TolA and Maximal Size Consistency"
+
+In accordance with <B>(1)</B> the Maximal Size for the Model is [by default]:
+
+<CENTER><B><PRE>Smax = 1.e-7 / 1.e-12 = 1.e+5  (2)</PRE></B></CENTER>
+
+\section sec2 Minimal Size of the Model
+
+The Minimal Size of the Model is defined as maximal diameter of enclosed
+sphere built for the smallest BRep entity of the Model.
+
+All models in Open CASCADE Technology are represented using double precision
+floating point numbers. This representation contains approximately 14-16
+significant digits.
+
+From the experience of using it is considered that the least four significant
+digits contain rounding-off errors occurring during the computation. So
+(taking in account the worst cases), there are ten reliable significant digits
+for double precision floating point numbers. Having the estimation it is
+possible to compute the value of the Minimal size of the model:
+
+<CENTER><B><PRE>Smin=Smax / 1.e+10             (3)</PRE></B></CENTER>
+
+In accordance with <B>(2)</B> for the default value it will be [by default]:
+
+<CENTER><B><PRE>Smin=1.e-5.                    (4)</PRE></B></CENTER>
+
+\section sec3 Full Range of Sizes
+
+The values <B>Smax (2)</B>, <B>Smin (4)</B> are theoretical. Taking into
+account the practical purposes of improving the reliability the lower limit
+should be restricted by one order. Thus, the full Range of Sizes of the Models
+is:
+
+<CENTER><B><PRE>[Smin, Smax] = [1.e-4, 1.e+5]  (5)</PRE></B></CENTER>
+
+*/