diff --git a/apps/readest-app/src/__tests__/components/rsvp-overlay-context.test.tsx b/apps/readest-app/src/__tests__/components/rsvp-overlay-context.test.tsx
index e4eac9ae..0c6485e5 100644
--- a/apps/readest-app/src/__tests__/components/rsvp-overlay-context.test.tsx
+++ b/apps/readest-app/src/__tests__/components/rsvp-overlay-context.test.tsx
@@ -343,6 +343,55 @@ describe('RSVPOverlay — CJK reading options', () => {
   });
 });
 
+describe('RSVPOverlay — RTL word display (#4630)', () => {
+  afterEach(() => {
+    cleanup();
+    localStorage.clear();
+  });
+
+  test('renders an Arabic word as a single RTL whole-word span, never split', () => {
+    const state = buildState({
+      words: [{ text: 'علم', orpIndex: 0, pauseMultiplier: 1 }],
+      currentIndex: 0,
+    });
+    const { container } = renderOverlay(state);
+
+    // Splitting the word into before/orp/after spans breaks Arabic shaping and
+    // reverses the visual order — RTL words must render whole instead.
+    const whole = container.querySelector('.rsvp-word-whole');
+    expect(whole).not.toBeNull();
+    expect(whole!.textContent).toBe('علم');
+    expect(whole!.getAttribute('dir')).toBe('rtl');
+    expect(container.querySelector('.rsvp-word-orp')).toBeNull();
+    expect(container.querySelector('.rsvp-word-before')).toBeNull();
+    expect(container.querySelector('.rsvp-word-after')).toBeNull();
+  });
+
+  test('renders a Hebrew word whole as well', () => {
+    const state = buildState({
+      words: [{ text: 'שלום', orpIndex: 0, pauseMultiplier: 1 }],
+      currentIndex: 0,
+    });
+    const { container } = renderOverlay(state);
+
+    const whole = container.querySelector('.rsvp-word-whole');
+    expect(whole).not.toBeNull();
+    expect(whole!.textContent).toBe('שלום');
+    expect(container.querySelector('.rsvp-word-orp')).toBeNull();
+  });
+
+  test('keeps the focus-letter split for Latin words (no spurious dir)', () => {
+    const state = buildState({
+      words: [{ text: 'hello', orpIndex: 1, pauseMultiplier: 1 }],
+      currentIndex: 0,
+    });
+    const { container } = renderOverlay(state);
+
+    expect(container.querySelector('.rsvp-word-orp')).not.toBeNull();
+    expect(container.querySelector('.rsvp-word-whole')).toBeNull();
+  });
+});
+
 describe('RSVPOverlay — manual word stepping (#4476)', () => {
   afterEach(() => cleanup());
 
diff --git a/apps/readest-app/src/__tests__/services/rsvp-utils.test.ts b/apps/readest-app/src/__tests__/services/rsvp-utils.test.ts
index 64d14fcd..f3af587d 100644
--- a/apps/readest-app/src/__tests__/services/rsvp-utils.test.ts
+++ b/apps/readest-app/src/__tests__/services/rsvp-utils.test.ts
@@ -3,6 +3,7 @@ import {
   isCJK,
   containsCJK,
   isCJKPunctuation,
+  isRTLText,
   getSegmenterLocale,
   segmentCJKText,
   splitTextIntoWords,
@@ -118,6 +119,41 @@ describe('rsvp/utils', () => {
     });
   });
 
+  describe('isRTLText', () => {
+    test('returns true for Arabic text', () => {
+      expect(isRTLText('علم')).toBe(true);
+      expect(isRTLText('مرحبا')).toBe(true);
+    });
+
+    test('returns true for Hebrew text', () => {
+      expect(isRTLText('שלום')).toBe(true);
+    });
+
+    test('returns true for Arabic presentation forms', () => {
+      expect(isRTLText('ﺍ')).toBe(true); // Arabic letter alef isolated form
+    });
+
+    test('returns true for text mixing Arabic and Latin', () => {
+      expect(isRTLText('Hello علم')).toBe(true);
+    });
+
+    test('returns false for pure Latin text', () => {
+      expect(isRTLText('hello')).toBe(false);
+    });
+
+    test('returns false for CJK text', () => {
+      expect(isRTLText('你好世界')).toBe(false);
+    });
+
+    test('returns false for digits and Latin punctuation', () => {
+      expect(isRTLText('123, 456.')).toBe(false);
+    });
+
+    test('returns false for empty string', () => {
+      expect(isRTLText('')).toBe(false);
+    });
+  });
+
   describe('getSegmenterLocale', () => {
     test('returns ja for Japanese hiragana text', () => {
       expect(getSegmenterLocale('こんにちは')).toBe('ja');
diff --git a/apps/readest-app/src/app/reader/components/rsvp/RSVPOverlay.tsx b/apps/readest-app/src/app/reader/components/rsvp/RSVPOverlay.tsx
index 328276ed..24970260 100644
--- a/apps/readest-app/src/app/reader/components/rsvp/RSVPOverlay.tsx
+++ b/apps/readest-app/src/app/reader/components/rsvp/RSVPOverlay.tsx
@@ -4,7 +4,7 @@ import React, { useEffect, useState, useCallback, useRef, useMemo } from 'react'
 import clsx from 'clsx';
 import { Insets } from '@/types/misc';
 import { RsvpState, RSVPController } from '@/services/rsvp';
-import { containsCJK } from '@/services/rsvp/utils';
+import { containsCJK, isRTLText } from '@/services/rsvp/utils';
 import { useThemeStore } from '@/store/themeStore';
 import { useSettingsStore } from '@/store/settingsStore';
 import { TOCItem } from '@/libs/document';
@@ -351,6 +351,10 @@ const RSVPOverlay: React.FC<RSVPOverlayProps> = ({
   const orpChar = currentWord ? currentWord.text.charAt(currentWord.orpIndex) : '';
   const wordAfter = currentWord ? currentWord.text.substring(currentWord.orpIndex + 1) : '';
   const isCJKWord = currentWord ? containsCJK(currentWord.text) : false;
+  // RTL words (Arabic, Hebrew, …) must never be split into before/orp/after
+  // spans: slicing by character index breaks letter shaping and reverses the
+  // visual order. Render them whole instead, like CJK Highlight Word (#4630).
+  const isRTLWord = currentWord ? isRTLText(currentWord.text) : false;
   const wordLetterSpacing = undefined;
   const wordSideOffset = isCJKWord ? '0.45em' : '0.3em';
 
@@ -973,12 +977,16 @@ const RSVPOverlay: React.FC<RSVPOverlayProps> = ({
                 }}
               >
                 {currentWord ? (
-                  isCJKWord && highlightWholeWord ? (
-                    // Whole-word mode: center the full CJK word and color every
-                    // character, instead of anchoring a single focus character.
+                  isRTLWord || (isCJKWord && highlightWholeWord) ? (
+                    // Whole-word mode: center the full word and color it, instead
+                    // of anchoring a single focus character. Used for CJK Highlight
+                    // Word and always for RTL words, whose shaping/order would
+                    // break if sliced into before/orp/after spans (#4630). dir=rtl
+                    // restores correct letter order and connection for RTL.
                     <span
                       className='rsvp-word-whole relative z-10 font-bold'
                       style={{ color: effectiveOrpColor }}
+                      dir={isRTLWord ? 'rtl' : undefined}
                     >
                       {currentWord.text}
                     </span>
diff --git a/apps/readest-app/src/services/rsvp/utils.ts b/apps/readest-app/src/services/rsvp/utils.ts
index 43ed7404..a08256bf 100644
--- a/apps/readest-app/src/services/rsvp/utils.ts
+++ b/apps/readest-app/src/services/rsvp/utils.ts
@@ -34,6 +34,20 @@ export function containsCJK(text: string): boolean {
   return false;
 }
 
+// Strong right-to-left scripts: Hebrew, Arabic, Syriac, Thaana, N'Ko,
+// Samaritan, Mandaic and the Arabic Extended blocks (U+0590–U+08FF), plus the
+// Hebrew/Arabic presentation forms (U+FB1D–U+FDFF and U+FE70–U+FEFF).
+const RTL_PATTERN = /[\u0590-\u08FF\uFB1D-\uFDFF\uFE70-\uFEFF]/;
+
+/**
+ * Check if text contains any right-to-left characters. Splitting such text by
+ * character index (as the RSVP focus-letter layout does) breaks letter shaping
+ * and reverses the visual order, so RTL words are rendered whole instead (#4630).
+ */
+export function isRTLText(text: string): boolean {
+  return RTL_PATTERN.test(text);
+}
+
 /**
  * Check if text is CJK punctuation
  */