diff --git a/CHANGES b/CHANGES
index 5f09616908..13f0353171 100644
--- a/CHANGES
+++ b/CHANGES
@@ -27,6 +27,10 @@ jsoup changelog
     memoizing the </title> scan and reducing GC.
     <https://github.com/jhy/jsoup/issues/1644>
 
+  * Improvement: when parsing custom tags (in HTML or XML), added a flyweight cache on Tag.valueOf(name) to reduce
+    memory overhead when many tags are repeated.
+
+
   * Bugfix: when tracking errors or checking for validity in the Cleaner, errors were incorrectly raised for missing
     optional closing tags.
 
diff --git a/src/main/java/org/jsoup/parser/HtmlTreeBuilder.java b/src/main/java/org/jsoup/parser/HtmlTreeBuilder.java
index aaf16dbf55..1c82827713 100644
--- a/src/main/java/org/jsoup/parser/HtmlTreeBuilder.java
+++ b/src/main/java/org/jsoup/parser/HtmlTreeBuilder.java
@@ -131,7 +131,7 @@ List<Node> parseFragment(String inputFragment, @Nullable Element context, String
                 default:
                     tokeniser.transition(TokeniserState.Data);
             }
-            root = new Element(Tag.valueOf(contextTag, settings), baseUri);
+            root = new Element(tagFor(contextTag, settings), baseUri);
             doc.appendChild(root);
             stack.add(root);
             resetInsertionMode();
@@ -245,13 +245,13 @@ Element insert(final Token.StartTag startTag) {
             return el;
         }
 
-        Element el = new Element(Tag.valueOf(startTag.name(), settings), null, settings.normalizeAttributes(startTag.attributes));
+        Element el = new Element(tagFor(startTag.name(), settings), null, settings.normalizeAttributes(startTag.attributes));
         insert(el);
         return el;
     }
 
     Element insertStartTag(String startTagName) {
-        Element el = new Element(Tag.valueOf(startTagName, settings), null);
+        Element el = new Element(tagFor(startTagName, settings), null);
         insert(el);
         return el;
     }
@@ -262,7 +262,7 @@ void insert(Element el) {
     }
 
     Element insertEmpty(Token.StartTag startTag) {
-        Tag tag = Tag.valueOf(startTag.name(), settings);
+        Tag tag = tagFor(startTag.name(), settings);
         Element el = new Element(tag, null, settings.normalizeAttributes(startTag.attributes));
         insertNode(el);
         if (startTag.isSelfClosing()) {
@@ -277,7 +277,7 @@ Element insertEmpty(Token.StartTag startTag) {
     }
 
     FormElement insertForm(Token.StartTag startTag, boolean onStack, boolean checkTemplateStack) {
-        Tag tag = Tag.valueOf(startTag.name(), settings);
+        Tag tag = tagFor(startTag.name(), settings);
         FormElement el = new FormElement(tag, null, settings.normalizeAttributes(startTag.attributes));
         if (checkTemplateStack) {
             if(!onStack("template"))
diff --git a/src/main/java/org/jsoup/parser/HtmlTreeBuilderState.java b/src/main/java/org/jsoup/parser/HtmlTreeBuilderState.java
index 4074cca349..e437cd9813 100644
--- a/src/main/java/org/jsoup/parser/HtmlTreeBuilderState.java
+++ b/src/main/java/org/jsoup/parser/HtmlTreeBuilderState.java
@@ -891,7 +891,7 @@ else if (!tb.onStack(formatEl)) {
                     } else if (node == formatEl)
                         break;
 
-                    Element replacement = new Element(Tag.valueOf(node.nodeName(), ParseSettings.preserveCase), tb.getBaseUri());
+                    Element replacement = new Element(tb.tagFor(node.nodeName(), ParseSettings.preserveCase), tb.getBaseUri());
                     // case will follow the original node (so honours ParseSettings)
                     tb.replaceActiveFormattingElement(node, replacement);
                     tb.replaceOnStack(node, replacement);
diff --git a/src/main/java/org/jsoup/parser/TreeBuilder.java b/src/main/java/org/jsoup/parser/TreeBuilder.java
index e21d70c586..f895c9ed28 100644
--- a/src/main/java/org/jsoup/parser/TreeBuilder.java
+++ b/src/main/java/org/jsoup/parser/TreeBuilder.java
@@ -10,7 +10,9 @@
 import javax.annotation.ParametersAreNonnullByDefault;
 import java.io.Reader;
 import java.util.ArrayList;
+import java.util.HashMap;
 import java.util.List;
+import java.util.Map;
 
 /**
  * @author Jonathan Hedley
@@ -24,6 +26,7 @@ abstract class TreeBuilder {
     protected String baseUri; // current base uri, for creating new elements
     protected Token currentToken; // currentToken is used only for error tracking.
     protected ParseSettings settings;
+    protected Map<String, Tag> seenTags; // tags we've used in this parse; saves tag GC for custom tags.
 
     private Token.StartTag start = new Token.StartTag(); // start tag to process
     private Token.EndTag end  = new Token.EndTag();
@@ -44,6 +47,7 @@ protected void initialiseParse(Reader input, String baseUri, Parser parser) {
         currentToken = null;
         tokeniser = new Tokeniser(reader, parser.getErrors());
         stack = new ArrayList<>(32);
+        seenTags = new HashMap<>();
         this.baseUri = baseUri;
     }
 
@@ -57,6 +61,7 @@ Document parse(Reader input, String baseUri, Parser parser) {
         reader = null;
         tokeniser = null;
         stack = null;
+        seenTags = null;
 
         return doc;
     }
@@ -159,4 +164,13 @@ protected void error(String msg, Object... args) {
     protected boolean isContentForTagData(String normalName) {
         return false;
     }
+
+    protected Tag tagFor(String tagName, ParseSettings settings) {
+        Tag tag = seenTags.get(tagName); // note that we don't normalize the cache key. But tag via valueOf may be normalized.
+        if (tag == null) {
+            tag = Tag.valueOf(tagName, settings);
+            seenTags.put(tagName, tag);
+        }
+        return tag;
+    }
 }
diff --git a/src/main/java/org/jsoup/parser/XmlTreeBuilder.java b/src/main/java/org/jsoup/parser/XmlTreeBuilder.java
index 5fad99e2c7..97ef372c37 100644
--- a/src/main/java/org/jsoup/parser/XmlTreeBuilder.java
+++ b/src/main/java/org/jsoup/parser/XmlTreeBuilder.java
@@ -83,7 +83,7 @@ private void insertNode(Node node) {
     }
 
     Element insert(Token.StartTag startTag) {
-        Tag tag = Tag.valueOf(startTag.name(), settings);
+        Tag tag = tagFor(startTag.name(), settings);
         // todo: wonder if for xml parsing, should treat all tags as unknown? because it's not html.
         if (startTag.hasAttributes())
             startTag.attributes.deduplicate(settings);
diff --git a/src/test/java/org/jsoup/parser/XmlTreeBuilderTest.java b/src/test/java/org/jsoup/parser/XmlTreeBuilderTest.java
index 8cd6bdd53c..c254e8e9cf 100644
--- a/src/test/java/org/jsoup/parser/XmlTreeBuilderTest.java
+++ b/src/test/java/org/jsoup/parser/XmlTreeBuilderTest.java
@@ -298,4 +298,20 @@ public void handlesLTinScript() {
         String out = doc.html();
         assertEquals("<body style=\"color: red\" name=\"\"><div></div></body>", out);
     }
+
+    @Test void customTagsAreFlyweights() {
+        String xml = "<foo>Foo</foo><foo>Foo</foo><FOO>FOO</FOO><FOO>FOO</FOO>";
+        Document doc = Jsoup.parse(xml, Parser.xmlParser());
+        Elements els = doc.children();
+
+        Tag t1 = els.get(0).tag();
+        Tag t2 = els.get(1).tag();
+        Tag t3 = els.get(2).tag();
+        Tag t4 = els.get(3).tag();
+        assertEquals("foo", t1.getName());
+        assertEquals("FOO", t3.getName());
+        assertSame(t1, t2);
+        assertSame(t3, t4);
+
+    }
 }