diff --git a/ddlmod.go b/ddlmod.go
index 9c93e6a..c839cd7 100644
--- a/ddlmod.go
+++ b/ddlmod.go
@@ -17,23 +17,11 @@ var (
 	indexRegexp        = regexp.MustCompile(fmt.Sprintf(`(?is)CREATE(?: UNIQUE)? INDEX [%v]?[\w\d-]+[%v]?(?s:.*?)ON (.*)$`, sqliteSeparator, sqliteSeparator))
 	tableRegexp        = regexp.MustCompile(fmt.Sprintf(`(?is)(CREATE TABLE [%v]?[\w\d-]+[%v]?)(?:\s*\((.*)\))?`, sqliteSeparator, sqliteSeparator))
 	separatorRegexp    = regexp.MustCompile(fmt.Sprintf("[%v]", sqliteSeparator))
-	columnsRegexp      = regexp.MustCompile(fmt.Sprintf(`[(,][%v]?(\w+)[%v]?`, sqliteSeparator, sqliteSeparator))
 	columnRegexp       = regexp.MustCompile(fmt.Sprintf(`^[%v]?([\w\d]+)[%v]?\s+([\w\(\)\d]+)(.*)$`, sqliteSeparator, sqliteSeparator))
 	defaultValueRegexp = regexp.MustCompile(`(?i) DEFAULT \(?(.+)?\)?( |COLLATE|GENERATED|$)`)
 	regRealDataType    = regexp.MustCompile(`[^\d](\d+)[^\d]?`)
 )
 
-func getAllColumns(s string) []string {
-	allMatches := columnsRegexp.FindAllStringSubmatch(s, -1)
-	columns := make([]string, 0, len(allMatches))
-	for _, matches := range allMatches {
-		if len(matches) > 1 {
-			columns = append(columns, matches[1])
-		}
-	}
-	return columns
-}
-
 type ddl struct {
 	head    string
 	fields  []string
@@ -110,9 +98,10 @@ func parseDDL(strs ...string) (*ddl, error) {
 				if strings.HasPrefix(fUpper, "CONSTRAINT") {
 					matches := uniqueRegexp.FindStringSubmatch(f)
 					if len(matches) > 0 {
-						if columns := getAllColumns(matches[1]); len(columns) == 1 {
+						cols, err := parseAllColumns(matches[1])
+						if err == nil && len(cols) == 1 {
 							for idx, column := range result.columns {
-								if column.NameValue.String == columns[0] {
+								if column.NameValue.String == cols[0] {
 									column.UniqueValue = sql.NullBool{Bool: true, Valid: true}
 									result.columns[idx] = column
 									break
@@ -123,12 +112,15 @@ func parseDDL(strs ...string) (*ddl, error) {
 					continue
 				}
 				if strings.HasPrefix(fUpper, "PRIMARY KEY") {
-					for _, name := range getAllColumns(f) {
-						for idx, column := range result.columns {
-							if column.NameValue.String == name {
-								column.PrimaryKeyValue = sql.NullBool{Bool: true, Valid: true}
-								result.columns[idx] = column
-								break
+					cols, err := parseAllColumns(f)
+					if err == nil {
+						for _, name := range cols {
+							for idx, column := range result.columns {
+								if column.NameValue.String == name {
+									column.PrimaryKeyValue = sql.NullBool{Bool: true, Valid: true}
+									result.columns[idx] = column
+									break
+								}
 							}
 						}
 					}
diff --git a/ddlmod_parse_all_columns.go b/ddlmod_parse_all_columns.go
new file mode 100644
index 0000000..760acf8
--- /dev/null
+++ b/ddlmod_parse_all_columns.go
@@ -0,0 +1,117 @@
+package sqlite
+
+import (
+	"errors"
+	"fmt"
+)
+
+type parseAllColumnsState int
+
+const (
+	parseAllColumnsState_NONE parseAllColumnsState = iota
+	parseAllColumnsState_Beginning
+	parseAllColumnsState_ReadingRawName
+	parseAllColumnsState_ReadingQuotedName
+	parseAllColumnsState_EndOfName
+	parseAllColumnsState_State_End
+)
+
+func parseAllColumns(in string) ([]string, error) {
+	s := []rune(in)
+	columns := make([]string, 0)
+	state := parseAllColumnsState_NONE
+	quote := rune(0)
+	name := make([]rune, 0)
+	for i := 0; i < len(s); i++ {
+		switch state {
+		case parseAllColumnsState_NONE:
+			if s[i] == '(' {
+				state = parseAllColumnsState_Beginning
+			}
+		case parseAllColumnsState_Beginning:
+			if isSpace(s[i]) {
+				continue
+			}
+			if isQuote(s[i]) {
+				state = parseAllColumnsState_ReadingQuotedName
+				quote = s[i]
+				continue
+			}
+			if s[i] == '[' {
+				state = parseAllColumnsState_ReadingQuotedName
+				quote = ']'
+				continue
+			} else if s[i] == ')' {
+				return columns, fmt.Errorf("unexpected token: %s", string(s[i]))
+			}
+			state = parseAllColumnsState_ReadingRawName
+			name = append(name, s[i])
+		case parseAllColumnsState_ReadingRawName:
+			if isSeparator(s[i]) {
+				state = parseAllColumnsState_Beginning
+				columns = append(columns, string(name))
+				name = make([]rune, 0)
+				continue
+			}
+			if s[i] == ')' {
+				state = parseAllColumnsState_State_End
+				columns = append(columns, string(name))
+			}
+			if isQuote(s[i]) {
+				return nil, fmt.Errorf("unexpected token: %s", string(s[i]))
+			}
+			if isSpace(s[i]) {
+				state = parseAllColumnsState_EndOfName
+				columns = append(columns, string(name))
+				name = make([]rune, 0)
+				continue
+			}
+			name = append(name, s[i])
+		case parseAllColumnsState_ReadingQuotedName:
+			if s[i] == quote {
+				// check if quote character is escaped
+				if i+1 < len(s) && s[i+1] == quote {
+					name = append(name, quote)
+					i++
+					continue
+				}
+				state = parseAllColumnsState_EndOfName
+				columns = append(columns, string(name))
+				name = make([]rune, 0)
+				continue
+			}
+			name = append(name, s[i])
+		case parseAllColumnsState_EndOfName:
+			if isSpace(s[i]) {
+				continue
+			}
+			if isSeparator(s[i]) {
+				state = parseAllColumnsState_Beginning
+				continue
+			}
+			if s[i] == ')' {
+				state = parseAllColumnsState_State_End
+				continue
+			}
+			return nil, fmt.Errorf("unexpected token: %s", string(s[i]))
+		case parseAllColumnsState_State_End:
+			break
+		}
+	}
+	if state != parseAllColumnsState_State_End {
+		return nil, errors.New("unexpected end")
+	}
+	return columns, nil
+}
+
+func isSpace(r rune) bool {
+	return r == ' ' || r == '\t'
+}
+
+func isQuote(r rune) bool {
+	return r == '`' || r == '"' || r == '\''
+}
+
+func isSeparator(r rune) bool {
+	return r == ','
+}
diff --git a/ddlmod_parse_all_columns_test.go b/ddlmod_parse_all_columns_test.go
new file mode 100644
index 0000000..eb70cdd
--- /dev/null
+++ b/ddlmod_parse_all_columns_test.go
@@ -0,0 +1,48 @@
+package sqlite
+
+import "testing"
+
+func TestParseAllColumns(t *testing.T) {
+	tc := []struct {
+		name     string
+		input    string
+		expected []string
+	}{
+		{
+			name:     "Simple case",
+			input:    "PRIMARY KEY (column1, column2)",
+			expected: []string{"column1", "column2"},
+		},
+		{
+			name:     "Quoted column name",
+			input:    "PRIMARY KEY (`column,xxx`, \"column 2\", \"column)3\", 'column''4', \"column\"\"5\")",
+			expected: []string{"column,xxx", "column 2", "column)3", "column'4", "column\"5"},
+		},
+		{
+			name:     "Japanese column name",
+			input:    "PRIMARY KEY (カラム1, `カラム2`)",
+			expected: []string{"カラム1", "カラム2"},
+		},
+		{
+			name:     "Column name quoted with []",
+			input:    "PRIMARY KEY ([column1], [column2])",
+			expected: []string{"column1", "column2"},
+		},
+	}
+	for _, tt := range tc {
+		t.Run(tt.name, func(t *testing.T) {
+			cols, err := parseAllColumns(tt.input)
+			if err != nil {
+				t.Errorf("Failed to parse columns: %s", err)
+			}
+			if len(cols) != len(tt.expected) {
+				t.Errorf("Expected %d columns, got %d", len(tt.expected), len(cols))
+			}
+			for i, col := range cols {
+				if col != tt.expected[i] {
+					t.Errorf("Expected %s, got %s", tt.expected[i], col)
+				}
+			}
+		})
+	}
+}