Home → Guides → Choosing a method

CFOP, Roux or ZZ?

Guide · about 9 minutes

This question gets asked constantly and answered badly, usually by someone defending the method they already use. Here is the honest version, including the part where the answer is boring.

The short answer

All three are fast enough that the method is not what is limiting you. There are sub-7 solvers using CFOP and sub-7 solvers using Roux. There are excellent ZZ solvers. At every level below world class, the difference between these methods is dwarfed by the difference between practising well and practising badly.

So if you are looking for the method that will make you fast: it does not exist, and the search for it is one of the more effective ways to waste a year.

That said, they genuinely differ, and the differences matter for what practice feels like and where your ceiling sits in specific sub-disciplines.

CFOP

Cross → F2L → OLL → PLL. Solve a cross on one face, insert four corner-edge pairs to complete the first two layers, orient the last layer, permute it.

What it is good at

CFOP is intuitive where it matters and algorithmic where that helps. F2L — the bulk of the solve — is intuitive pair-solving that rewards lookahead directly, which means practice translates into improvement in a very legible way. The last layer is pure pattern recognition and execution, which is trainable in isolation.

It is also, by a wide margin, the best-documented method in existence. Every case has a dozen tutorials, every algorithm has alternatives, and every plateau you hit has been written about. When you get stuck, help is available in a way it simply is not for the others.

And it is the most common method at competitions, so the person sitting next to you can probably help.

What it is not good at

CFOP is move-inefficient. A typical CFOP solve is somewhere around 55-60 moves. It compensates with speed: the algorithms are heavily optimised for finger tricks, and turning fast is easier than turning cleverly.

It also relies on cube rotations during F2L, which cost time and break lookahead. Good solvers minimise them; nobody eliminates them.

And the last layer is a lot of memorisation — 78 algorithms for full OLL and PLL, and considerably more if you go further. See the guide on learning them for how long that actually takes.

Roux

First block → second block → corners of the last layer → last six edges. Build a 1x2x3 block on the left, a matching one on the right, orient and permute the four top corners, then finish the remaining six edges with M and U moves.

What it is good at

Roux is move-efficient — typically around 45-50 moves, meaningfully fewer than CFOP. It achieves this with block-building, which is far more flexible than a cross: there are many ways to build a block, so you can nearly always find something efficient in the scramble in front of you.

It requires very few algorithms. The corner step is a small set, and the final edge step is largely intuitive. You can be genuinely fast on a fraction of the memorisation CFOP demands.

It barely rotates. Roux solves are mostly R, M and U, and the absence of rotations helps lookahead considerably.

And it is exceptional one-handed. The M-slice-heavy final step suits one hand better than CFOP's last layer does, which is why Roux is disproportionately represented among top one-handed solvers.

What it is not good at

Block-building is harder to learn and harder to teach. There is no case list. Getting good at it means developing judgement about the scramble, which is a slower and less legible process than learning algorithms — you improve, but you cannot always point at what improved.

The M-slice work needs a cube that handles M moves well, and it needs finger technique most CFOP solvers never develop.

There is far less material available, and fewer people to ask.

It also transfers poorly to big cubes. 4x4 and up are usually solved by reduction to a 3x3, and that reduction lands you in a state that CFOP handles more naturally. Most Roux solvers use CFOP on big cubes, which means learning both anyway.

ZZ

EOLine → F2L → last layer. Orient all twelve edges while placing two cross edges, then build the first two layers with no F or B moves at all, then finish the last layer.

What it is good at

The idea is genuinely elegant. Once edges are oriented, the entire rest of the solve is R, U and L moves — ergonomically the fastest moves on the cube. No F or B, no rotations during F2L.

Because edges arrive pre-oriented, the last layer is simpler: you skip edge orientation entirely, and with the appropriate algorithm sets you can finish in one look more readily than in CFOP.

Move counts sit between CFOP and Roux, and the turning is fast because of the restricted move set.

What it is not good at

EOLine is hard. You must work out edge orientation for all twelve edges during a 15-second inspection, and it is genuinely difficult — a real skill that takes months to become reliable rather than a step you learn in an afternoon.

It is the least popular of the three by a wide margin, so material and community are thinner still than for Roux.

And getting the full benefit means learning large algorithm sets beyond standard OLL/PLL, which is a substantial commitment on top of an already demanding first step.

Comparison

So which one

If you are starting out: CFOP. Not because it is best, but because the material, the community and the structured path are worth more than a ten-move efficiency difference you cannot yet exploit. You can always switch later, and most people who "should" have started with Roux do fine.

If you dislike memorisation, or one-handed is your event: Roux. These are real reasons rather than aesthetic ones. Roux genuinely asks less of your memory and genuinely suits one hand better.

If you like the intellectual side and are patient: ZZ. Go in knowing EOLine will be frustrating for months and that you will be figuring things out with less help than the others get.

The part people do not want to hear

Switching methods is almost always the wrong move.

The pattern is familiar: someone plateaus at 20 seconds on CFOP, reads that Roux is more efficient, switches, and is back at 45 seconds. Six months later they are at 22 — slower than when they started, having spent half a year rebuilding what they already had.

Plateaus are almost never caused by the method. They are caused by lookahead, by turning technique, by recognition speed, by practising a hundred casual solves instead of focused ones. Every one of those transfers across methods, which is exactly why switching does not fix them — you take the problem with you.

Before switching, be honest about a few things. Is your F2L genuinely smooth, or do you pause between every pair? Do you look ahead, or look-turn-look-turn? Are you practising deliberately, or just solving? Is your last layer fluent under pressure, or only when relaxed?

If any answer is uncomfortable, that is your bottleneck, and it will still be your bottleneck in a new method.

What to do instead

Whatever method you use, the things that actually make you faster are the same: slow deliberate solves focused on never pausing, drilling recognition separately from execution, and practising under conditions that resemble the ones you care about.

If competitions are the goal, that last one matters most — pressure attacks lookahead first, and lookahead is method-independent. Run a simulated round and see what your average does when the attempts count. That number tells you more about what to work on than any method comparison will.