diff --git a/packages/rich-text/src/normalise-formats.js b/packages/rich-text/src/normalise-formats.js
index 349540df9dc256..24690f089b4c76 100644
--- a/packages/rich-text/src/normalise-formats.js
+++ b/packages/rich-text/src/normalise-formats.js
@@ -17,7 +17,7 @@ export function normaliseFormats( { formats, text, start, end } ) {
 	newFormats.forEach( ( formatsAtIndex, index ) => {
 		const lastFormatsAtIndex = newFormats[ index - 1 ];
 
-		if ( lastFormatsAtIndex ) {
+		if ( lastFormatsAtIndex && formatsAtIndex ) {
 			const newFormatsAtIndex = formatsAtIndex.slice( 0 );
 
 			newFormatsAtIndex.forEach( ( format, formatIndex ) => {
@@ -29,6 +29,8 @@ export function normaliseFormats( { formats, text, start, end } ) {
 			} );
 
 			newFormats[ index ] = newFormatsAtIndex;
+		} else if ( ! formatsAtIndex ) {
+			delete newFormats[ index ];
 		}
 	} );
 
diff --git a/packages/rich-text/src/test/normalise-formats.js b/packages/rich-text/src/test/normalise-formats.js
index 8984bcb17771b8..f0a1b827d264d5 100644
--- a/packages/rich-text/src/test/normalise-formats.js
+++ b/packages/rich-text/src/test/normalise-formats.js
@@ -13,19 +13,36 @@ import { getSparseArrayLength } from './helpers';
 describe( 'normaliseFormats', () => {
 	const strong = { type: 'strong' };
 	const em = { type: 'em' };
+	const record = deepFreeze( {
+		formats: [ , [ em ], [ { ...em }, { ...strong } ], [ em, strong ], , , ],
+		text: 'one two',
+	} );
 
-	it( 'should normalise formats', () => {
-		const record = {
-			formats: [ , [ em ], [ { ...em }, { ...strong } ], [ em, strong ] ],
-			text: 'one two three',
-		};
-		const result = normaliseFormats( deepFreeze( record ) );
-
-		expect( result ).toEqual( record );
-		expect( result ).not.toBe( record );
+	function isNormalised( result ) {
 		expect( getSparseArrayLength( result.formats ) ).toBe( 3 );
 		expect( result.formats[ 1 ][ 0 ] ).toBe( result.formats[ 2 ][ 0 ] );
 		expect( result.formats[ 1 ][ 0 ] ).toBe( result.formats[ 3 ][ 0 ] );
 		expect( result.formats[ 2 ][ 1 ] ).toBe( result.formats[ 3 ][ 1 ] );
+	}
+
+	it( 'should normalise formats', () => {
+		const result = normaliseFormats( record );
+
+		expect( result ).toEqual( record );
+		expect( result ).not.toBe( record );
+		isNormalised( result );
+	} );
+
+	it( 'should be serializable', () => {
+		const result = normaliseFormats( record );
+		expect( JSON.stringify( record ) ).toBe( JSON.stringify( result ) );
+	} );
+
+	it( 'should normalise serialized formats', () => {
+		const deserialized = JSON.parse( JSON.stringify( record ) );
+		const result = normaliseFormats( deserialized );
+
+		expect( result ).toEqual( record );
+		isNormalised( result );
 	} );
 } );
diff --git a/packages/rich-text/src/test/to-html-string.js b/packages/rich-text/src/test/to-html-string.js
index a65c3cb772447d..eaef8f2f651b62 100644
--- a/packages/rich-text/src/test/to-html-string.js
+++ b/packages/rich-text/src/test/to-html-string.js
@@ -78,4 +78,15 @@ describe( 'toHTMLString', () => {
 
 		expect( toHTMLString( create( { element } ) ) ).toEqual( expectedHTML );
 	} );
+
+	it( 'should be able to work from serialized value (without references)', () => {
+		expect( toHTMLString( {
+			text: 'abc',
+			formats: [
+				null,
+				[ { type: 'strong' } ],
+				null,
+			],
+		} ) ).toEqual( 'a<strong>b</strong>c' );
+	} );
 } );
diff --git a/packages/rich-text/src/to-tree.js b/packages/rich-text/src/to-tree.js
index c1ced44a16c6b1..0681de01e35f27 100644
--- a/packages/rich-text/src/to-tree.js
+++ b/packages/rich-text/src/to-tree.js
@@ -3,6 +3,7 @@
  */
 
 import { split } from './split';
+import { normaliseFormats } from './normalise-formats';
 
 export function toTree( value, multilineTag, settings ) {
 	if ( multilineTag ) {
@@ -35,7 +36,7 @@ export function toTree( value, multilineTag, settings ) {
 		onEndIndex,
 		onEmpty,
 	} = settings;
-	const { formats, text, start, end } = value;
+	const { formats, text, start, end } = normaliseFormats( value );
 	const formatsLength = formats.length + 1;
 	const tree = createEmpty( tag );