SAA-C03 Study Guide: What the Exam Covers, How Much Each Domain Weighs, and a Study Order That Works
The AWS Certified Solutions Architect – Associate exam (SAA-C03) is less about remembering service names and more about picking the right design when a scenario gives you two or three constraints at once. This guide explains what the exam covers, how the four domains are weighted, the order in which we suggest studying them, and the traps that cost people the most points.
Study guide · AWS Solutions Architect – Associate (SAA-C03)

The exam at a glance
| Questions | 65 (50 scored, 15 unscored that are not marked as such) |
| Time | 130 minutes, about 2 minutes per question |
| Format | Multiple choice (one answer) and multiple response (two or more answers) |
| Passing score | 720 on a scaled score of 100–1,000 |
Two minutes per question sounds generous until you meet a stem of six or seven lines with a table of requirements. Plan to answer the short definition questions in under a minute and bank that time for the long scenarios.
The four domains and what each one really tests
| Domain | Weight | What it really asks |
|---|---|---|
| Design Secure Architectures | 30% | Who can reach what: IAM policies and roles, multi-account guardrails, network boundaries, encryption and key ownership. |
| Design Resilient Architectures | 26% | What happens when a part fails: Multi-AZ, decoupling with queues, recovery objectives (RPO and RTO) and disaster recovery patterns. |
| Design High-Performing Architectures | 24% | Matching storage, compute, database and network choices to an access pattern: caching, read replicas, the right storage type. |
| Design Cost-Optimized Architectures | 20% | Paying only for what the requirement needs: purchase options, storage classes and lifecycle rules, data transfer paths. |
The weights are close to each other, so no domain can be skipped. Security carries the most weight and also shows up inside the other three: a "resilient" or "cost" question often has one option that quietly breaks a security requirement.
What the questions look like
We wrote 395 original SAA-C03 questions for this site, following the published exam guide. Looking at them as a whole shows the shape of the exam better than any single question:
- About two in three are scenarios ("A company runs… it needs… which solution…"). Only about one in ten asks for a plain definition.
- About one in twelve asks you to select two answers. Both must be right; there is no partial credit.
- Amazon S3 appears more than any other service, followed by RDS, IAM, VPC, EC2 and DynamoDB. If one service deserves extra hours, it is S3: storage classes, lifecycle rules, bucket policies, encryption options, replication and access points.
A study order that works
Studying in the order of the exam guide is not the fastest route, because the domains lean on each other. This order builds each layer on the one before it:
- Networking and identity first (1–2 weeks). VPC, subnets, route tables, security groups versus network ACLs, VPC endpoints; IAM users, roles, policies and the policy evaluation logic. Nearly every scenario sits on top of these.
- Storage and databases (1–2 weeks). S3 in depth, EBS versus EFS versus FSx, RDS and Aurora (Multi-AZ, read replicas, global databases), DynamoDB (capacity modes, global tables, DAX), ElastiCache.
- Compute and decoupling (1 week). EC2 purchase options, Auto Scaling, load balancers, Lambda, containers on ECS and Fargate, SQS, SNS and EventBridge.
- Resilience and performance patterns (1 week). Backup and restore, pilot light, warm standby and multi-site; CloudFront, Global Accelerator and Route 53 routing policies.
- Cost last (a few days). Once you know the services, cost questions become a matter of reading the requirement carefully: Savings Plans versus Reserved Instances versus Spot, S3 Intelligent-Tiering, data transfer paths.
Beginner tip: if VPC terms blur together, build one by hand once: a VPC with one public and one private subnet, an internet gateway, a NAT gateway and two route tables. Launch an instance in each subnet and try to reach the internet from both. Twenty minutes of clicking teaches what route tables do better than any diagram, and it is what half of the networking questions describe in words. Delete the NAT gateway afterwards: it is billed by the hour.
After each block, do a few practice sets for the domain you just studied, then one mixed set. Mixing early stops you from recognising the domain instead of reading the question.
The traps that cost the most points
1. Missing the qualifier at the end of the stem
"Most cost-effective", "least operational overhead", "most secure" and "with minimal changes to the application" each point to a different correct answer for the same architecture. Several options usually work; only one matches the qualifier. Read the last sentence of the stem first, then the scenario.
Beginner tip: keep a small two-column list while you practise: on the left the qualifier ("least operational overhead"), on the right what it pointed to in the questions you reviewed ("managed or serverless service"). After three or four sets the list stops growing, and you will recognise the pattern before you finish reading the options.
2. Treating high availability and scaling as the same thing
An RDS Multi-AZ deployment keeps a standby for failover; it does not serve reads (for the classic one-standby setup). Read replicas add read capacity and can be promoted, but replication to them is asynchronous. When a stem says "survive the loss of an Availability Zone", think Multi-AZ; when it says "reporting queries slow down the primary", think read replicas.
Going deeper: why the standby cannot serve reads, and when it can
In a classic RDS Multi-AZ DB instance, every write is replicated synchronously to a standby in another Availability Zone before it is acknowledged. The standby exists only to take over: it has no endpoint you can query, and a failover (typically one to two minutes) moves the DNS name to it. That is why it gives you no read capacity, and why it loses no committed data.
Two newer designs blur the line. A Multi-AZ DB cluster (MySQL and PostgreSQL) has one writer and two standbys in different AZs that do serve reads, with faster failover. Aurora separates compute from storage: the cluster volume keeps six copies across three AZs and needs four of them to confirm a write, and Aurora Replicas read from that same volume, so a replica is both read capacity and a failover target. If a question says "Aurora" and "high availability" together, an Aurora Replica in another AZ is usually the answer, not a separate standby.
3. Choosing a working answer that is more work
Running Kafka on EC2, writing a cron job on an instance or building a custom replication script often works. When a managed service does the same job (MSK, EventBridge Scheduler, S3 Replication), the managed one is almost always the expected answer, especially with "least operational overhead".
4. Forgetting minimum storage durations
Moving objects to S3 Standard-IA, Glacier Flexible Retrieval or Glacier Deep Archive looks cheaper, but those classes charge for a minimum storage duration (30, 90 and 180 days) and for retrieval. For data that is read often or deleted soon, the "cheaper" class can cost more.
5. Mixing up gateway and interface endpoints
Private access from a VPC to S3 or DynamoDB without a NAT gateway is the textbook case for a gateway endpoint, which has no hourly charge. Interface endpoints (PrivateLink) serve most other services and are billed per hour and per GB.
In the last week
- Do at least two full mixed sets under time pressure, and review every option of every question, including the ones you got right.
- Write down each mistake as a rule ("Multi-AZ is not a read scaling feature") rather than as a fact about one question.
- Re-read the qualifiers in your wrong answers. If most mistakes come from the last sentence of the stem, slow down there, not everywhere.
Practice by domain
Every SAA-C03 question in the bank explains each option, right and wrong. Start with the domain you just studied, then switch to mixed practice and a timed mock exam.
- SAA-C03 exam page · All 495 SAA-C03 questions · Timed mock exam · 28-day study plan
- By domain: Design Secure Architectures · Design Resilient Architectures · Design High-Performing Architectures · Design Cost-Optimized Architectures
Exam facts as published in the AWS exam guide for SAA-C03; check the official guide before you book, as AWS updates it from time to time.