What the AWS Certified Machine Learning Engineer – Associate exam actually asks.

• 65 questions in 170 minutes — the scored format in the exam guide (50 scored plus 15 unscored), which is the shape this room practises. The beta currently delivers 85 questions in the same 170 minutes, so the room's timer is deliberately generous against that paper.
• It sits at the Associate level and assumes hands-on work with Amazon SageMaker AI and Amazon Bedrock, not model research.
• 50 questions are scored and 15 are unscored trial items, so a few questions will not count. You are not told which.
• 720 out of 1000 to pass, on a scaled score. Because it is scaled, the raw number of correct answers you need is not fixed or published.
• Question formats: multiple choice and multiple response. A multiple-response item tells you how many to select and expects exactly that many.
• Answer everything. A blank scores nothing, so flag the long ones, bank the quick ones, and return before time runs out.
• $75 USD during the beta; AWS lists $150 USD as the standard price and the beta is English-only. MLA-C02 is the current edition of the exam.

Two hours fifty is enough for sixty-five scenario questions only if you keep moving. Decide early how long a long scenario may take before you flag it and move on.

• Domain 1 — Data Preparation for ML and AI (28%). The largest domain. Collecting and storing data, choosing formats and storage by access pattern, streaming ingestion, feature engineering and the feature store, preparing data for foundation models, and validating quality and bias.
• Domain 2 — ML Model and Foundation Model Development (24%). Choosing between a custom model, a managed AI service, retrieval and fine-tuning; training and tuning; customising foundation models; and evaluating predictive and generative systems.
• Domain 3 — Deployment and Orchestration of ML and AI Workflows (24%). The four inference shapes, capacity and provisioning, containers and private networking, and the pipelines and gates that take a model from a training job to production.
• Domain 4 — Operating, Monitoring, and Securing ML and AI Solutions (24%). Data and model quality monitoring, drift, generative-AI observability, cost and latency, and least-privilege access to models, data and endpoints.

The four domains are close in weight, so there is no safe domain to skip. Domain 1 is only four points larger than the rest, and the three at 24% are each worth about sixteen questions.

Expect a short scenario — a nightly scoring job, a retrieval assistant, a latency-budgeted endpoint, a pipeline that must be reproducible — followed by a question about the best design or the next change. More than one option may work in isolation; the exam asks for the one that fits every stated constraint. The guide's task statements name the skills each domain tests, and AWS publishes them in full, so confirm them there before you sit the paper.

Read the last sentence first, then the scenario, then the options. Qualifiers to look for:

• "Nightly" or "millions of records" → Batch transform, not an always-on endpoint.
• "Labels arrive days later" → Monitor inputs and drift now; do not claim you can measure accuracy.
• "Cannot leave the VPC" → A VPC endpoint and a private path, not a public service call with a narrower role.
• "Must not add operational burden" → A managed capability before a self-managed cluster.

The questions test operating judgement, so knowing that a service exists is not enough. The study that pays is being able to say which service fits a constraint and why the near neighbour does not: batch transform against asynchronous inference, data quality against model quality, retrieval against fine-tuning, a managed knowledge base against a self-managed index.

Three habits that work:

• Build one pipeline end to end. Data in S3, a processing job, a training job, a registered model, an endpoint and a monitor touch most of the domains and make the questions concrete.
• Read the AWS documentation alongside the guide. SageMaker AI and Amazon Bedrock each have their own guide; the exam guide names the topics but the documentation settles the mechanism.
• Review every option, not just your answer. On this exam several options are plausible; learning why one breaks a stated constraint is the skill being tested.

MockRoom's MLA-C02 study room has five distinct 65-question mock exams — 325 original scenario questions in total, written against the published exam guide and weighted across its four domains, each with an explanation for every option. It also has 104 active recall flashcards for spaced review. Nothing here is taken from an exam dump or from the guide's own sample questions, and none of it is real examination content.

Open the MLA-C02 study room