diff --git a/names.go b/names.go
new file mode 100644
index 0000000..227f126
--- /dev/null
+++ b/names.go
@@ -0,0 +1,146 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package hujson
+
+import (
+	"reflect"
+	"strings"
+)
+
+// Names is a map of canonical JSON object names.
+// The value for each entry is another map of JSON object names to use
+// for any JSON sub-objects.
+//
+// As a special case, a map with only a single entry where the key is "*"
+// indicates that the sub-map of names is to be applied to all sub-objects.
+//
+// See the example for Value.NormalizeNames for more information.
+type Names map[string]Names
+
+// NewNames constructs a Names map for the provided type
+// as typically understood by the "encoding/json" package.
+//
+// See the example for Value.NormalizeNames for more information.
+func NewNames(t reflect.Type) Names {
+	// TODO(dsnet): Handle cycles in the type graph.
+	// TODO(dsnet): What happens when t implements json.Unmarshaler?
+	switch t.Kind() {
+	case reflect.Array, reflect.Slice, reflect.Ptr:
+		return NewNames(t.Elem())
+	case reflect.Map:
+		names := NewNames(t.Elem())
+		if len(names) == 0 {
+			return nil
+		}
+		return Names{"*": names}
+	case reflect.Struct:
+		names := make(Names)
+		for i := 0; i < t.NumField(); i++ {
+			sf := t.Field(i)
+			if sf.PkgPath != "" {
+				// TODO(dsnet): Technically, an embedded, unexported type with
+				// exported fields can have serializable fields.
+				// This almost never occurs in practice.
+				continue // unexported fields are ignored
+			}
+
+			// Derive JSON name from either the Go field name or `json` tag.
+			name := sf.Name
+			inlined := sf.Anonymous && mayIndirect(sf.Type).Kind() == reflect.Struct
+			switch tag := sf.Tag.Get("json"); tag {
+			case "":
+				break // do nothing
+			case "-":
+				continue // explicitly ignored field
+			default:
+				if i := strings.IndexByte(tag, ','); i >= 0 {
+					tag = tag[:i]
+				}
+				if tag != "" {
+					name = tag
+					inlined = false // explicitly named fields are never inlined
+				}
+			}
+
+			// If inlined, hoist all child names up to the parent.
+			// Otherwise, just insert the current name.
+			if inlined {
+				// TODO(dsnet): This does not properly handle name conflicts.
+				// However, conflicts rarely occur in practice.
+				// See https://github.com/golang/go/blob/aa4e0f528e1e018e2847decb549cfc5ac07ecf20/src/encoding/json/encode.go#L1352-L1378
+				for name, subNames := range NewNames(sf.Type) {
+					names[name] = subNames
+				}
+			} else {
+				names[name] = NewNames(sf.Type)
+			}
+		}
+		if len(names) == 0 {
+			return nil
+		}
+		return names
+	default:
+		return nil
+	}
+}
+
+func mayIndirect(t reflect.Type) reflect.Type {
+	if t.Kind() == reflect.Ptr {
+		t = t.Elem()
+	}
+	return t
+}
+
+// NormalizeNames recursively iterates through v and replaces any JSON object
+// names that is a case-insensitive match with a name found in names,
+// with the canonical name found in names.
+//
+// See the example for Value.NormalizeNames for more information.
+func (v *Value) NormalizeNames(names Names) {
+	v.normalizeNames(names)
+	v.UpdateOffsets()
+}
+func (v *Value) normalizeNames(names Names) {
+	if len(names) == 0 {
+		return
+	}
+	switch v2 := v.Value.(type) {
+	case *Object:
+		// If names is a map with only a "*" key,
+		// then apply the same subNames map to all map values.
+		if subNames, ok := names["*"]; ok && len(names) == 1 {
+			for i := range v2.Members {
+				v2.Members[i].Value.normalizeNames(subNames)
+			}
+			break
+		}
+
+		for i := range v2.Members {
+			name := v2.Members[i].Name.Value.(Literal).String()
+
+			// Fast-path: Exact match with names map.
+			subNames, ok := names[name]
+			if !ok {
+				// Slow-path: Case-insensitive match with names map.
+				var match string
+				for name2 := range names {
+					if (match == "" || match < name2) && strings.EqualFold(name, name2) {
+						match = name2
+					}
+				}
+				// If a case-insensitive match was found, update the name.
+				if match != "" {
+					v2.Members[i].Name.Value = String(match)
+					subNames = names[match]
+				}
+			}
+			v2.Members[i].Value.normalizeNames(subNames)
+		}
+	case *Array:
+		for i := range v2.Elements {
+			v2.Elements[i].normalizeNames(names)
+		}
+	}
+}
diff --git a/names_test.go b/names_test.go
new file mode 100644
index 0000000..27aafb0
--- /dev/null
+++ b/names_test.go
@@ -0,0 +1,186 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package hujson
+
+import (
+	"fmt"
+	"log"
+	"reflect"
+	"testing"
+
+	"github.com/google/go-cmp/cmp"
+)
+
+// The "encoding/json" package unfortunately uses case-insensitive matching
+// when unmarshaling. For example, the following:
+//
+//	{"NAME": ...}
+//	{"nAmE": ...}
+//	{"name": ...}
+//	{"Name": ...}
+//
+// are all equivalent when unmarshaling into a Go struct like:
+//
+//	struct{ Name string }
+//
+// In order to conform some HuJSON value to consistently use the same set of
+// JSON object names, a Names map can be derived from Go struct type
+// and applied upon the HuJSON value using the Value.NormalizeNames method.
+func ExampleValue_NormalizeNames() {
+	type MyStruct struct {
+		Alpha int
+		Bravo []struct {
+			Foo int
+		} `json:"bravo_wavo"`
+		Charlie map[string]struct {
+			Fizz int `json:"fizzy_wizzy"`
+			Buzz int `json:",omitempty"`
+		}
+		Ignored    int `json:"-"`
+		unexported int
+	}
+
+	// Derive the set of canonical names from the Go struct type.
+	names := NewNames(reflect.TypeOf(MyStruct{}))
+	// Verify that the derived names match what we expect.
+	gotNames := names
+	wantNames := Names{
+		"Alpha": nil, // name comes from Go struct field
+		"bravo_wavo": { // name comes from `json` tag
+			"Foo": nil, // name comes from Go struct field
+		},
+		"Charlie": { // name comes from Go struct field
+			"*": { // implies that all JSON object members use the same set of sub-names
+				"fizzy_wizzy": nil, // name comes from `json` tag
+				"Buzz":        nil, // name comes from Go struct field
+			},
+		},
+	}
+	if diff := cmp.Diff(gotNames, wantNames); diff != "" {
+		log.Fatalf("NewNames mismatch (-want +got):\n%s", diff)
+	}
+
+	// Parse some HuJSON input with strangely formatted names.
+	v, err := Parse([]byte(`{
+	"AlPhA": 0,
+	"BRAVO_WAVO": [
+		{"FOO": 0},
+		{"fOo": 1},
+		{"Foo": 2},
+	],
+	"charlie": {
+		"kEy": {"FIZZY_WIZZY": 0},
+		"KeY": {"bUzZ": 1},
+	},
+}`))
+	if err != nil {
+		log.Fatal(err)
+	}
+	// Conform JSON object names in the HuJSON value to the canonical names.
+	v.NormalizeNames(gotNames)
+	fmt.Println(v)
+
+	// Output:
+	// {
+	// 	"Alpha": 0,
+	// 	"bravo_wavo": [
+	// 		{"Foo": 0},
+	// 		{"Foo": 1},
+	// 		{"Foo": 2},
+	// 	],
+	// 	"Charlie": {
+	// 		"kEy": {"fizzy_wizzy": 0},
+	// 		"KeY": {"Buzz": 1},
+	// 	},
+	// }
+}
+
+func TestNormalizeNames(t *testing.T) {
+	type MyStruct struct {
+		GoName   int
+		JSONName int `json:"json_name"`
+	}
+
+	tests := []struct {
+		typ       interface{}
+		wantNames Names
+		in        string
+		wantOut   string
+	}{{
+		typ:       0,
+		wantNames: nil,
+		in:        `{"hello":"goodbye"}`,
+		wantOut:   `{"hello":"goodbye"}`,
+	}, {
+		typ:       new(int),
+		wantNames: nil,
+		in:        `{"hello":"goodbye"}`,
+		wantOut:   `{"hello":"goodbye"}`,
+	}, {
+		typ: struct {
+			GoName1    int
+			GoName2    int `json:",omitempty"`
+			JSONName   int `json:"json_name"`
+			Ignored    int `json:"-"`
+			unexported int `json:"fake_name"`
+		}{},
+		wantNames: Names{"GoName1": nil, "GoName2": nil, "json_name": nil},
+		in:        `{"goname1":0,"goname2":0,"JSON_NAME":0,"JSONNAME":0}`,
+		wantOut:   `{"GoName1":0,"GoName2":0,"json_name":0,"JSONNAME":0}`,
+	}, {
+		typ: struct {
+			M *[]map[int][]map[string][]struct {
+				F int `json:"field"`
+			}
+		}{},
+		wantNames: Names{"M": {"*": {"*": {"field": nil}}}},
+		in:        `{"m":[{"hello":[{"goodbye":[{"FIELD":0}]}]}]}`,
+		wantOut:   `{"M":[{"hello":[{"goodbye":[{"field":0}]}]}]}`,
+	}, {
+		typ: struct {
+			M map[string]struct{}
+		}{},
+		wantNames: Names{"M": nil},
+	}, {
+		typ: struct {
+			MyStruct
+			int
+		}{},
+		wantNames: Names{"GoName": nil, "json_name": nil},
+	}, {
+		typ: struct {
+			*MyStruct
+			*int
+		}{},
+		wantNames: Names{"GoName": nil, "json_name": nil},
+	}, {
+		typ: struct {
+			MyStruct `json:"my_struct"`
+		}{},
+		wantNames: Names{"my_struct": {"GoName": nil, "json_name": nil}},
+	}}
+
+	for _, tt := range tests {
+		t.Run("", func(t *testing.T) {
+			gotNames := NewNames(reflect.TypeOf(tt.typ))
+			if diff := cmp.Diff(tt.wantNames, gotNames); diff != "" {
+				t.Errorf("NewNames(%T) mismatch (-want +got):\n%s", tt.typ, diff)
+			}
+
+			if tt.in == "" {
+				return
+			}
+			v, err := Parse([]byte(tt.in))
+			if err != nil {
+				t.Fatalf("Parse error: %v", err)
+			}
+			v.NormalizeNames(gotNames)
+			gotOut := v.String()
+			if diff := cmp.Diff(tt.wantOut, gotOut); diff != "" {
+				t.Errorf("v.NormalizeNames(%T) mismatch (-want +got):\n%s", tt.typ, diff)
+			}
+		})
+	}
+}