Selenium Interview Questions for 3 Years Experience (2026): 40+ Real Questions

The client sends commands over the W3C WebDriver protocol directly to the browser driver (chromedriver, geckodriver). No JSON Wire Protocol bridge, no Selenium Server in the local flow.

W3C standardisation, relative locators, BiDi support, improved Grid 4 (Docker/K8s native), built-in CDP access.

findElement throws NoSuchElementException when nothing matches; findElements returns an empty list.

Stable test IDs (data-testid) > CSS > XPath. Avoid index-based XPath; it breaks on minor DOM changes.

WebElement email = driver.findElement(
  with(By.tagName("input")).below(By.id("label-email"))
);

Wait on a stable attribute, use contains() / starts-with(), or anchor on a nearby stable element with a relative locator.

Implicit: global, applies to every findElement. Explicit: condition-based per call (WebDriverWait). Fluent: explicit + polling interval + ignored exceptions. Never mix implicit and explicit — Selenium docs warn against it.

elementToBeClickable, visibilityOfElementLocated, presenceOfElementLocated, invisibilityOf, textToBePresentInElement.

Wait on a network-completion signal (a spinner disappearing, a specific element appearing) — never Thread.sleep.

Expected answer: language + runner (Java + TestNG/JUnit5), POM + PageFactory or component-based pages, config via .properties / .yaml, data via JSON/Excel, parallel execution, Allure/ExtentReports, Maven/Gradle + GitHub Actions/Jenkins.

POM: classic, one class per page. Screenplay: actor-centric, scales for complex flows. Component: pages composed of reusable widgets — best for modern SPAs.

External JSON/Excel for static data, faker libraries for dynamic data, API setup/teardown for state — never hard-code.

Each test sets up its own data via APIs, runs in its own browser session, and cleans up after itself.

TestNG: parallel="methods" in suite XML. JUnit 5: junit.jupiter.execution.parallel.enabled=true. Ensure WebDriver is thread-local (ThreadLocal<WebDriver>).

Distributed execution — Hub + Nodes (or fully distributed). Supports Docker and Kubernetes deployment out of the box.

Dockerised nodes, autoscaled on Kubernetes, or hosted on BrowserStack/Sauce Labs. Match concurrency to your CI runner count.

Reproduce locally with retries off, capture screenshot + HTML snapshot + browser console + network HAR on failure, check for race conditions and bad waits, fix root cause — never blindly retry.

Only for known-environmental flakes (network blips). Track retry counts; if a test retries >3% of runs, it's broken, not flaky.

• Stable selectors (data-testid).
• Explicit waits on real signals.
• API-driven setup, not UI clicks.
• One browser session per test.
• Quarantine flaky tests in a separate suite.

Native sendKeys on the file input — no AutoIT/Robot needed for simple inputs.

Configure browser to a known download dir, then assert the file exists with the expected size/hash.

Pass credentials in the URL (https://user:pass@host) or use Chrome DevTools Protocol (Network.setExtraHTTPHeaders) in Selenium 4.

Capture driver.getWindowHandles(), iterate, switchTo().window(handle).

driver.switchTo().frame(...) — by index, name or WebElement. Always switch back with defaultContent().

Selenium 4: element.getShadowRoot().findElement(By.cssSelector(...)).

DevTools dt = ((ChromeDriver) driver).getDevTools();
dt.createSession();
dt.send(Network.enable(...));

Faster setup, deterministic state, fewer flaky UI steps. Use RestAssured/RestSharp for setup/teardown — see our RestSharp guide and the API testing questions.

Use JDBC/JPA to query the DB after a UI action and assert state — but keep it for critical flows only.

Headless Chrome/Firefox in GitHub Actions or Jenkins, Maven/Gradle build, results published as JUnit XML + Allure HTML artifact, Slack/Teams notifications on failure.

Tag-based execution (TestNG groups / JUnit 5 @Tag), separate pipeline jobs per tag, regression nightly + smoke on every PR.

Profile slowest tests, parallelise across 4–8 threads, move setup to APIs, kill UI smoke duplication, shard across CI runners.

Browser version mismatch, headless rendering quirks, timezone/locale, missing env vars, race conditions exposed by slower CI machines. Reproduce with the exact CI Docker image.

Explain ROI: automate stable, high-value flows; keep exploratory and visual checks manual. 100% is a vanity metric.

Educate on data-testid, add a PR check that fails if locators change without coordination, contribute the testids yourself.

Run full regression nightly, smoke on PRs, parallelise, use spot instances or BrowserStack quota smartly.

For all of these, structure answers as Situation → Action → Result → Learning. Rehearse them out loud in the AI Mock Interview — reading the answer isn't enough.