XML to XSD

updated 26 August 2026

What can and cannot be inferred from one document

A schema states what is allowed. A document only shows what occurred. Everything interesting about generating one from the other lives in that gap, so here is exactly which way each guess is made:

• Repeating elements. maxOccurs="unbounded" is only asserted when a repeat was actually seen — two or more siblings of the same name inside one parent. A list with a single entry in your sample produces maxOccurs="1", which is almost certainly too tight; that is the first thing to widen by hand.
• Optional elements and attributes. minOccurs="0" and use="optional" are asserted when some instance of the parent had the child and another did not. With only one instance, everything present looks required.
• Types widen, never narrow. A value is typed as the narrowest built-in every observed value satisfies, checked in order: boolean, integer, decimal, dateTime, date, time, anyURI, then string. One non-numeric value among a thousand numbers makes the whole element xs:string, which is the safe direction to be wrong in.
• Child order. xs:sequence is a claim that children always appear in that order. If two instances of the same parent listed their children differently, that claim is false, so the model becomes xs:all instead and the status line says how many elements it happened to. Override it if you know the real rule.
• Mixed content. An element with both child elements and its own non-whitespace text gets mixed="true", because without it a validator rejects the text.
• Attributes on text-only elements need xs:simpleContent with an xs:extension — a detail that is easy to get wrong by hand and produces a schema that silently rejects your document.

Nested or flat

Nested uses anonymous inline types, so the schema reads in the same shape as the document. It is the easier one to check against a sample by eye, and the right default.

Flat declares every element globally and refers to them by ref. That is the shape you want if the schema will be maintained by hand or imported elsewhere, because each element is defined exactly once — but it also means two elements that happen to share a name must share a definition, so their models are merged.

Either way, an element name that appears in several places is modelled once from every occurrence combined. If your document uses <item> for two genuinely different things, the inferred type is the union of both and will be looser than either. Rename one, or split the schema.

Namespaces

If the sample's root is in a namespace, that becomes the targetNamespace and elementFormDefault="qualified" is set, which is what makes an unprefixed child element in your document validate. A document in no namespace produces a no-namespace schema. Multiple namespaces in one document are beyond what a single inferred schema can express — the first is used and the rest are kept as literal prefixed names, which will need an xs:import adding by hand.

FAQ