diff --git a/docs/guides/query-selectors.md b/docs/guides/query-selectors.md
new file mode 100644
index 00000000000..ef590d48d08
--- /dev/null
+++ b/docs/guides/query-selectors.md
@@ -0,0 +1,118 @@
+# Query Selectors
+
+Queries are the primary mechanism for interacting with the DOM on your site. For example, a typical workflow goes like:
+
+```ts
+// Import puppeteer
+import puppeteer from 'puppeteer';
+
+(async () => {
+  // Launch the browser
+  const browser = await puppeteer.launch();
+
+  // Create a page
+  const page = await browser.newPage();
+
+  // Go to your site
+  await page.goto('YOUR_SITE');
+
+  // Query for an element handle.
+  const element = await page.waitForSelector('div > .class-name');
+
+  // Do something with element...
+  await element.click(); // Just an example.
+
+  // Dispose of handle
+  await element.dispose();
+
+  // Close browser.
+  await browser.close();
+})();
+```
+
+## CSS
+
+CSS selectors follow the CSS spec of the browser being automated. We provide some basic type deduction for CSS selectors (such as `HTMLInputElement` for `input`), but any selector that contains no type information (such as `.class-name`) will need to be coerced manually using TypeScript's `as` coercion mechanism.
+
+### Example
+
+```ts
+// Automatic
+const element = await page.waitForSelector('div > input');
+// Manual
+const element = (await page.waitForSelector(
+  'div > .class-name-for-input'
+)) as HTMLInputElement;
+```
+
+## Built-in selectors
+
+Built-in selectors are Puppeteer's own class of selectors for doing things CSS cannot. Every built-in selector starts with a prefix `.../` to assist Puppeteer in distinguishing between CSS selectors and a built-in.
+
+### Text selectors (`text/`)
+
+Text selectors will select "minimal" elements containing the given text, even within (open) shadow roots. Here, "minimum" means the deepest elements that contain a given text, but not their parents (which technically will also contain the given text).
+
+#### Example
+
+```ts
+// Note we usually need type coercion since the type cannot be deduced, but for text selectors, `instanceof` checks may be better for runtime validation.
+const element = await page.waitForSelector('text/My name is Jun');
+```
+
+### XPath selectors (`xpath/`)
+
+XPath selectors will use the browser's native [`Document.evaluate`](https://developer.mozilla.org/en-US/docs/Web/API/Document/evaluate) to query for elements.
+
+#### Example
+
+```ts
+// There is not type deduction for XPaths.
+const node = await page.waitForSelector('xpath/h2');
+```
+
+### ARIA selectors (`aria/`)
+
+ARIA selectors can be used to find elements with a given ARIA label. These labels are computed using Chrome's internal representation.
+
+#### Example
+
+```ts
+const node = await page.waitForSelector('aria/Button name');
+```
+
+### Pierce selectors (`pierce/`)
+
+Pierce selectors will run the `querySelector*` API on the document and all shadow roots to find an element.
+
+:::danger
+
+Selectors will **not** _partially_ pierce through shadow roots. See the examples below.
+
+:::
+
+#### Example
+
+Suppose the HTML is
+
+```html
+<div>
+  <custom-element>
+    <div></div>
+  </custom-element>
+</div>
+```
+
+Then
+
+```ts
+// This will be two elements because of the outer and inner div.
+expect((await page.$$('pierce/div')).length).toBe(2);
+
+// Partial piercing doesn't work.
+expect((await page.$$('pierce/div div')).length).toBe(0);
+```
+
+## Custom selectors
+
+Puppeteer provides users the ability to add their own query selectors to Puppeteer using [Puppeteer.registerCustomQueryHandler](../api/puppeteer.registercustomqueryhandler.md). This is useful for creating custom selectors based on framework objects or other vendor-specific objects.