AWS Certified Machine Learning Engineer Associate (MLA-C01): Exam Domains and Weights

First, check which version your test date falls under

AWS is refreshing this certification. As of August 2026, MLA-C01 is the version you sit, but registration for the updated exam (MLA-C02) opens on September 1, 2026, and the last day to take MLA-C01 in English is September 28, 2026. The update brings generative AI, agentic AI, and foundation model workloads into scope alongside traditional ML engineering.

Everything below describes MLA-C01. If your test date lands after the changeover, download the MLA-C02 guide instead and treat the weights here as historical.

Exam format at a glance

All figures from the official AWS Certified Machine Learning Engineer Associate (MLA-C01) exam guide. The 15 unscored questions are not identified during the exam and do not affect your result.

The four content domains

The weights below are the share of scored content, so they describe how the 50 graded questions are distributed. They sit in a narrow band, from 22% to 28%, which is the single most useful thing to know about this exam: there is no small domain you can safely skim.

Domain 1: Data Preparation for Machine Learning (28%)

The largest domain, and it is data engineering more than data science. It covers ingesting and storing data for ML, transforming it and engineering features, and validating that the data is fit to train on. Expect the AWS data stack (S3, AWS Glue, Amazon EMR, Amazon Athena) alongside SageMaker data tooling such as Data Wrangler and Feature Store, plus data formats, labeling, class imbalance, and bias detection before a model exists.

Domain 2: ML Model Development (26%)

Choosing a modeling approach, training and refining models, and analyzing how they performed. This covers picking between built-in SageMaker algorithms, pre-trained models, and custom training, hyperparameter tuning, managing overfitting and underfitting, and reading evaluation metrics for classification and regression. It is the closest thing on the exam to a conventional ML knowledge check.

Domain 3: Deployment and Orchestration of ML Workflows (22%)

The smallest domain, and the one candidates from a data science background usually find least familiar. It covers selecting deployment infrastructure for a set of requirements (real-time endpoints, batch transform, serverless and asynchronous inference), sizing and scaling the compute behind it, provisioning that infrastructure as code, and wiring up CI/CD and orchestration with SageMaker Pipelines and related AWS services.

Domain 4: ML Solution Monitoring, Maintenance, and Security (24%)

Keeping a deployed solution healthy, affordable, and locked down. It covers monitoring inference quality and drift with SageMaker Model Monitor, observability through CloudWatch, cost and performance tuning of ML infrastructure, and securing ML resources with IAM, encryption, and network controls. This domain plus Domain 3 is 46% of the exam, which is why MLA-C01 reads as an engineering credential rather than a modeling one.

What the weights mean in actual questions

Percentages are easier to plan against once you convert them into question counts. With 50 scored questions, the four domains work out to roughly:

Counts are the published weights applied to 50 scored questions, rounded. AWS does not guarantee an exact per-domain count on any individual exam form.

How to use the weights when you study

On an exam with a 6-point spread between the largest and smallest domain, prioritising by weight barely helps. Prioritise by gap instead, and use the weights to decide when to stop, not where to start.

• Coming from data engineering? Domain 1 is already close to your day job. Domains 2 and 4 are where the runway goes: evaluation metrics, tuning, drift, and ML-specific security.
• Coming from data science? Domain 2 is your home ground and it is only a quarter of the exam. Domains 3 and 4, together 46%, are the ones that decide the result.
• Read the task statements. The exam guide lists task statements and in-scope services under each domain, and questions are written from them. Reading them straight through is the fastest blind-spot audit available, and it is free.
• Build one end-to-end solution. Training a model in SageMaker, deploying it to an endpoint, adding Model Monitor, and scripting the whole thing touches all four domains at once.

FAQ