CLRS 3rd vs 4th edition: what actually changed (and which one to buy)

If you're taking an algorithms course at MIT, Stanford, Berkeley, CMU, or basically any US CS program, the textbook on the syllabus is Introduction to Algorithms by Cormen, Leiserson, Rivest, and Stein — universally known as CLRS. The current edition (4th, 2022) lists for $139. The previous edition (3rd, 2009) routinely sells for $30-50 used.

The question every CS sophomore asks: do I need the 4th, or will the 3rd do?

Short answer: for almost everyone, the 3rd edition is fine. Here's the longer answer.

What's new in the 4th edition

MIT Press's official changelog identifies the substantive additions:

• 140 new exercises and 22 new problems — homework problem numbers shift
• New chapter on machine learning (Ch. 20) — covers gradient descent, basic neural network theory, regularization
• Heavily reworked dynamic programming chapter — now Ch. 14
• Expanded coverage of randomized algorithms in places
• New section on online algorithms in the linear programming chapter
• Cleaner notation in a few places (Θ and O are used more consistently)

The book grew from ~1,300 pages (3rd) to ~1,300 pages (4th) — most of the new content displaced older content, rather than padding length.

What's identical

The core 90% of the book — the curriculum every algorithms course actually uses — is essentially unchanged:

• Sorting (insertion, merge, heap, quicksort, counting/radix/bucket)
• Divide-and-conquer + master theorem
• Hash tables, BSTs, red-black trees
• Dynamic programming fundamentals (rod cutting, matrix-chain, LCS)
• Greedy algorithms
• Graph algorithms (BFS, DFS, MST, shortest paths, max flow)
• NP-completeness and reductions
• Approximation algorithms

If your professor's lectures cover these, the 3rd edition gives you 100% of what you need.

The "previous edition" trade-off

Buying the 3rd edition saves money. The trade-offs:

You lose:

• The new ML chapter (Ch. 20) — but if you're taking an ML course, you'll use Bishop PRML or Murphy's PML anyway, not CLRS.
• ~140 newer exercises — homework problem numbers won't match if your professor assigns specific problems by number.

You keep:

• Every theorem, every proof, every algorithm pseudocode that matters for the course.
• Roughly $90.

When you actually need the 4th edition

Three scenarios:

1. Your professor explicitly assigns homework by 4th-edition problem number. Check the syllabus. If problem numbers are referenced from the 4th edition, you need the 4th — the numbering shifted significantly between editions.
2. You're taking a graduate machine-learning-meets-algorithms course that specifically uses CLRS Ch. 20.
3. You're keeping the book long-term as a reference and want the most current edition for your shelf.

For an undergraduate algorithms course where lectures drive the homework? The 3rd is fine.

Where to buy each edition cheapest

Both editions are tracked on StembookDeals — click through to see live pricing across all five retailers we compare:

• CLRS 4th edition (9780262046305) — used as required reading in 9 canonical courses including MIT 6.006, Stanford CS 161, and Berkeley CS 170.
• CLRS 3rd edition (9780262033848) — still used as alternate/reference at most of those same schools.

The other classic algorithms books

If CLRS is too heavyweight or your course leans differently, two reasonable alternatives in our catalog:

• Skiena's The Algorithm Design Manual — more practical, less theoretical, very readable.
• Knuth TAOCP Volume 1 — for the historically-minded; probably not your textbook unless you're at Stanford specifically.

TL;DR

Save the $90. Buy CLRS 3rd edition used. The book that turns 17 this year still teaches algorithms fine.