Good eLearning course design starts with the job the learner must do, then teaches it in small pieces, with practice, feedback and review built in. Fifteen practices fit that description, and this guide rates the evidence behind each one.
Most advice on this topic is opinion. Several top-ranking guides give rules, such as slide caps, without a source. Here, each practice is tied to a dated source or marked as judgment, and graded.
You also get a grading table and a one-page pre-launch checklist. Both suit instructional designers, subject-matter experts (SMEs) who build their own courses, and teams that review others’ work.
Key takeaways
- Start from the job task and a measurable result. These steps rest on standards or judgment, not lab tests.
- The best-supported practices concern memory: retrieval practice, spacing, explanatory feedback and cutting extra material.
- Short videos, signaling and chunking help, but the evidence is smaller or less direct.
- Accessibility and a pilot with real learners are launch gates, not polish.
What does “hold up” mean in eLearning course design?
A practice holds up when repeated studies, a published standard or long practitioner use back it, and you can say where it fails. The grade tells you how confident to be, not how big the gain will be in your course.
The five grades are:
- Strong: many studies or meta-analyses agree.
- Moderate: research supports it, but effects are smaller or mixed.
- Limited: observational data only.
- Standard: a published standard or framework.
- Judgment: practitioner experience, with no direct test found.
Treat every number as a signal. Several studies below measure test scores or watch time, not job results.
How should you set up eLearning course design before building any screens?
Set the target before the content. Name the job behavior that must change, write objectives a person could observe, and find out what learners already know.
1. Start from the job task, not the topic
Action mapping, described on Cathy Moore’s blog, begins with a measurable business result and lists the observable behaviors people need on the job. It notes that training is “rarely the best or only solution.” The Kirkpatrick Model agrees: start with Level 4, the results you want. For a wider process, see instructional design models compared. Grade: judgment.
2. Write objectives a reviewer can observe
The 2001 revision of Bloom’s taxonomy, edited by Anderson and Krathwohl, says an objective has a verb and an object. The verb names the thinking, such as explain or apply. Vague verbs such as “know” cannot be checked. Our guide on how to write learning objectives has verb lists and rewrites. Grade: standard.
3. Check what learners already know
A 2003 review in Educational Psychologist described the expertise reversal effect. Techniques that help beginners can lose value, or even hurt, for experienced learners. Offer a skip-ahead check for mixed groups. Our piece on adult learning principles covers how prior experience shapes design. Grade: moderate.
How should you design each screen and video?
Give each screen one idea. Remove what does not serve it, pair words with fitting pictures, point to the key element, and keep segments short. Richard Mayer’s multimedia principles cover most of this, and his 2023 summary reports positive results in most tests.
4. Cut material that does not serve the objective
The coherence principle says to drop extra content and keep visuals that matter. In Mayer’s 2023 summary, 18 of 19 tests supported it, with a median effect size of 0.86. The figure comes from the principle’s own author. Cut decorative images and side stories. Grade: strong.
5. Pair words with pictures, and narrate instead of repeating text
Mayer reports a median effect size of 1.35 for words plus pictures over words alone, across 13 comparisons. Narration over on-screen text scores 1.00 in 18 of 19 tests. The redundancy result is weaker: 0.10, in 8 of 12 tests. So do not read a full paragraph aloud while it shows. Add captions (see practice 14). Grade: strong.
6. Signal what matters
Signaling means using color, contrast or cues to draw attention to key information. A 2020 meta-analysis of 29 experiments found a learning gain of d = 0.38. Mayer’s median is higher, at 0.70, so the sources disagree on size, not direction. Signal one thing per screen. Grade: moderate.
7. Break content into short, learner-paced segments
A 2019 meta-analysis of 56 investigations found small to medium segmenting effects on retention and transfer. Learners with high prior knowledge benefited more, so segmenting is not only for beginners. Segment by idea, not by the clock, and let learners set the pace. Grade: moderate.
8. Keep videos near six minutes or less
A 2014 study of edX course videos analyzed 6.9 million viewing sessions across 862 videos from four courses. Median engagement was at most six minutes, whatever the video length. Learners often watched under half of videos longer than nine minutes. The data are server logs from math and science courses, so treat six minutes as a starting point. For phones, see our responsive eLearning course guide. Grade: limited.
Pro tip: Cut First, Then Signal. Remove extras before you add emphasis. Then highlight only the one thing the learner must see on each screen. If three items glow, none of them stands out.
How should you teach the skill itself?
Teach the skill the way people will use it. Show a worked example, let learners practice a realistic task, then bring the idea back later. The strongest evidence in this guide sits here, in practice testing, spacing and feedback.
9. Show worked examples to novices, then fade them
A 2019 conference paper reviewing the research reports that novices learn more from worked examples than from minimal guidance. For experienced learners the pattern flips. Show a finished example, then a partly finished one, then a blank task. Grade: moderate.
10. Make practice look like the job
Merrill’s 2007 synthesis of instructional theories says to build instruction around real tasks with feedback that gradually fades. He says each principle has support in the works he reviewed, but gives no formal strength rating. Use short workplace situations, not recall-only questions. Grade: moderate.
11. Use retrieval practice instead of rereading
Dunlosky and colleagues reviewed ten study techniques in 2013. Practice testing and distributed practice earned the highest utility ratings. Rereading and highlighting rated low. In a course, that means low-stakes questions that make learners pull answers from memory. The review covered student study habits, so this is an inference. See our guide to knowledge check design. Grade: strong.
12. Space practice over time
A 2006 review in Psychological Bulletin found 839 assessments of spaced practice in 317 experiments. The best gap between sessions grew as the delay before the final test grew. Schedule recap questions days or weeks after the module, not on its last screen. The review covered verbal recall tasks, so test it on your own skills. Grade: strong.
13. Give feedback that explains
A 2020 meta-analysis of 435 studies found an overall feedback effect of d = 0.48. High-information feedback, which includes task and process detail, reached 0.99. Reinforcement or punishment reached 0.24. Results varied widely. Replace “Incorrect, try again” with why the answer fails and what to do next. Grade: strong.
Pro tip: Write the Feedback Before the Question. Draft the explanation for each wrong choice first. If you cannot say why an option is wrong, the question probably tests trivia, not the skill. This is practitioner advice.
What should you check before launching an eLearning course?
Check that every learner can use the course and that real learners can complete the task it teaches. Accessibility is a published standard, and a pilot finds problems you cannot see.
14. Build accessibility in from the first draft
WCAG 2.2 became a W3C Recommendation on December 12, 2024. It asks for captions on prerecorded audio in synchronized media (Level A) and a contrast ratio of at least 4.5:1 for normal text. At Level AA, pointer targets must be at least 24 by 24 CSS pixels, with listed exceptions. Fix these in the template. Whether a rule binds you depends on your client and jurisdiction, so ask an expert. Grade: standard.
15. Pilot with real learners and measure results, not just completion
Run the course with a small group from the real audience before launch. After launch, use the four Kirkpatrick levels: reaction, learning, behavior and results. Completion is not one of them. Plan those measures at the start, as in practice 1. Grade: judgment for the pilot, standard for the levels.
How strong is the evidence behind each practice?
Five practices grade strong, five moderate, one limited, two standard and two judgment. Memory-focused practices grade highest. Process practices grade lower, usually because they are harder to test.
| Practice | Grade | Basis | Main caveat |
|---|---|---|---|
| 1. Start from the job task | Judgment | Action mapping; Kirkpatrick Level 4 | Practitioner methods |
| 2. Observable objectives | Standard | 2001 Bloom revision | Framework, no effect size |
| 3. Check prior knowledge | Moderate | 2003 expertise reversal review | Needs a skip-ahead check |
| 4. Cut extras | Strong | Mayer 2023: 18 of 19 tests | Author’s own summary |
| 5. Words, pictures, narration | Strong | Mayer 2023: d = 1.35 and 1.00 | Redundancy gain only 0.10 |
| 6. Signal key points | Moderate | 29 experiments, d = 0.38 | Mayer’s median is 0.70 |
| 7. Short segments | Moderate | 56 investigations, small to medium | Gains larger with prior knowledge |
| 8. Short videos | Limited | edX logs, 862 videos | Observational logs |
| 9. Worked examples | Moderate | Studies on novices | Reverses for experts |
| 10. Job-like practice | Moderate | Merrill 2007 synthesis | No formal strength rating |
| 11. Retrieval practice | Strong | Dunlosky 2013 review | Student study techniques |
| 12. Spacing | Strong | 317 experiments, 2006 | Verbal recall tasks |
| 13. Explanatory feedback | Strong | 435 studies, d = 0.48 | High variation |
| 14. Accessibility | Standard | WCAG 2.2 | Legal duty varies |
| 15. Pilot and measure | Judgment | Kirkpatrick four levels | Pilot size is a guess |
These grades are our reading of the sources named above.
What goes on a one-page pre-launch review checklist?
Use this checklist as a final gate. It has fourteen yes-or-no questions drawn from the practices above, so a reviewer who did not build the course can run it. Fix each “no” or record why you accept it.
- Does the course name a job behavior and a measured result?
- Does every objective use an observable verb?
- Did you check prior knowledge or offer a skip-ahead?
- Does each screen carry one idea, with no decorative extras?
- Does narration add to the screen, not repeat it?
- Is one key element signaled on each screen?
- Are segments short and learner-paced, with videos near six minutes?
- Do novices see worked examples before open tasks?
- Does practice use realistic job situations?
- Are there recall questions after each section and again days later?
- Does feedback on wrong answers explain why?
- Are captions, contrast and touch targets checked against WCAG 2.2?
- Did a small group from the real audience try the course?
- Will you measure behavior and results, not just completion?
What should you do next?
Run the checklist on one existing course. Fix the nos that touch learning first: objectives, practice and feedback.
If you are building a new course, you can start free in SimpliAuthor. Based on the vendor’s published pages, it offers an editable canvas with text, media, interactive blocks and assessments, SCORM 1.2 or 2004 export, and AI-drafted outlines from your documents. The pages do not say what the free option includes, so check first.
SimpliAuthor is made by SimpliTrain, the same company that publishes this site, so weigh this suggestion accordingly. This guide does not test any authoring tool, and the product facts come from SimpliAuthor’s published pages as of November 2026.
Frequently asked questions
What are the main principles of eLearning course design?
Start from the job task, write observable objectives, remove extra material, pair words with fitting visuals, practice recall over time, give feedback that explains, and test with real learners. The sections above grade all fifteen practices by evidence.
How long should an eLearning course be?
No source in this guide sets one right length. Split content into short, learner-paced segments by idea. For video, one edX study found median engagement of six minutes or less. Let the objectives and the job task decide total course time.
How do you evaluate an eLearning course?
Use the four Kirkpatrick levels: reaction, learning, behavior and results. Decide first which results the course should change, then pick a measure for each level. Completion and satisfaction scores alone cover only a small part of that picture.
What makes an eLearning course accessible?
WCAG 2.2 sets the published checks. Examples include captions for prerecorded audio in synchronized media, a 4.5:1 contrast ratio for normal text, and pointer targets of at least 24 by 24 CSS pixels at Level AA. Build these into your template, and ask an expert which rules apply.
Do I need an authoring tool to apply these practices?
No. Objectives, practice design, feedback wording and the pre-launch checklist all happen before or outside any tool. The tool mainly affects how easily you can reuse templates, add captions and export SCORM. Start with a document or spreadsheet, and pick a tool once the design is settled.



