diff --git a/MODEL.md b/MODEL.md
new file mode 100644
index 00000000..a164d651
--- /dev/null
+++ b/MODEL.md
@@ -0,0 +1,103 @@
+# KDL Data Model
+
+This document specifies an abstract data model of KDL document language. 
+
+*This is version `0.0.0` of KDL Data model. It has not been released yet.*
+
+## Introduction
+
+KDL is defined by [KDL Specification](SPEC.md) as a formal language with
+components such as nodes, identifiers, strings, comments, and whitespace.
+Some of these components can be expressed in mutliple ways and some of
+these components are typically ignored when a KDL document is parsed.
+
+KDL Data model defines a conceptual structure of semantically relevant elements
+expressed in KDL syntax. The data model is required to process KDL documents
+independent from syntax and implementations with [KDL Query
+Language](QUERY-SPEC.md) and [KDL Schema Language](SCHEMA-SPEC.md).
+
+## Elements
+
+Every KDL document represents a [document](#document).
+
+Sequences and sets can be empty.
+
+A Unicode string is a sequence of Unicode code points.
+
+### Document
+
+A document is a sequence of [nodes](#node).
+
+Documents must not contain itself via node direct or indirect children to avoid
+circular structures.
+
+### Node
+
+A node consists of four mandatory and one optional elements:
+
+* a **name** being a Unicode string
+* an optional **tag** being a Unicode string (the empty string tag is distinct from no tag)
+* a sequence of **arguments**, each being a [value](#value)
+* a map of **properties**, each consisting of
+    * a key being a Unicode string unique within the map
+    * a [value](#value)
+* a sequence of **children** being a [document](#document)
+
+### Value
+
+A value consists of one of:
+
+* a Unicode string
+* a **number**, being an arbitrary-precision, base-10 decimal number value
+* a **boolean**, being one of the special values *true* and *false*
+* the special value *null*
+
+and
+
+* an optional a **tag** being a Unicode string (the empty string tag is distinct from no tag)
+
+The data model does not limit the use of tags to specific types of values.
+
+## Implementation Notes
+
+### Extensions to the data model
+
+While valid implementations must support at least the elements described above,
+they *may* recognize and preserve additional information not captured in the
+data model, such as:
+
+* Line numbers and character position of parsed KDL syntax elements.
+
+* Comments and precise details of whitespace and node terminators.
+
+* Whether a node had an empty child list (`node {}`) or no child list at all.
+  KDL syntax allows both. KDL data model considers these identical.
+ 
+* The precise format of numbers, such as what radix they're specified in
+  (`0x1a`), whether they are an integer or not (`1` vs `1.0`), the presence of
+  underscores (`1_234`), etc. KDL syntax supports multiple ways to specify
+  numbers. KDL data model does not differentiate number types.
+
+### Data binding
+
+The mapping of KDL elements to data elements of a particular programming or
+database languages is beyond the scope of this data model. Implementations
+should use tags as type annotations to map KDL data model instances to other
+type systems.
+
+### Limitations to the data model
+
+Implementations may choose to limit the set of processable KDL documents for
+technical reasons. Such limitations must be stated clearly to indicate a useful
+but incomplete support of KDL data model. Reasonable limitations include:
+
+* Precision of numbers
+
+* Types of elements that can have tags (e.g. disallow tags for boolean values
+  and `null`)
+
+* Unicode Normalization (e.g. collapse properties into one when their names are
+  equivalent after normalization)
+
+Implementations must document how limitations to KDL model are applied when KDL
+document are read (e.g. give warnings and ignore unsupported elements).