Plugin marketing has trained us to treat "saturation", "distortion" and "clipping" as three points on a single line that runs from subtle to extreme. The reality is two independent axes, and once you separate them, the question of which plugin to reach for stops being about taste and starts being about what you actually want the transfer function to do.
Distortion is the umbrella, not a separate flavour
Sound on Sound's working definition is the one to start from. Distortion is any nonlinear process that adds new frequency components to a signal. Saturation, overdrive, fuzz, and clipping are all flavours of distortion. They are not separate phenomena. When a plugin is labelled "distortion" it usually means the manufacturer wants you to hear the effect explicitly, while a "saturator" wants the effect to sit at the edge of audibility. That is a stylistic naming choice, not a technical one.
The two technical axes underneath those labels are transition sharpness and transfer-function symmetry.
Axis one: transition sharpness
How abruptly does the transfer function bend as the signal approaches the ceiling. A hard clipper has a discontinuous derivative right at the ceiling: the function jumps from "linear with slope 1" to "constant". A soft clipper smooths that corner with a polynomial or transcendental function. The classic textbook example is the cubic
f(x) = 1.5x - 0.5x^3documented in Julius Smith's PASP work at Stanford CCRMA. The cubic is chosen so the derivative is zero exactly at x = +/-1, which means the curve meets the flat clipping region with no kink. tanh(gx) is the other commonly used choice and behaves similarly.
Why does smoothness matter. Discontinuities in derivatives generate strong high-order harmonics. The Fourier series of a square wave, which is the extreme case of an instantaneous transition, contains odd harmonics that fall off only as 1/n. Soft transitions concentrate the energy in the first few harmonics and decay much faster. That is why a hard clipper sounds harsh and a soft saturator sounds polite even when both are removing the same number of dB from the peaks.
Axis two: symmetry
A transfer function f(x) that satisfies f(-x) = -f(x), so it has point symmetry about the origin, only produces odd harmonics from a sinusoidal input. This is provable from the Fourier series of f(sin(wt)). The even-harmonic coefficients integrate to zero. Functions like tanh, sign, x^3, a hard clipper, and any polynomial built only from odd-power terms all fall into this group.
To get even harmonics you need an asymmetric transfer function. Three common ways to introduce that asymmetry:
- Add a DC offset or low-frequency bias before a symmetric nonlinearity. The nonlinearity now sees an asymmetric input and the output contains even harmonics.
- Use an inherently asymmetric function, for example a soft knee on the positive side and a harder one on the negative side.
- Use a polynomial that contains even-power terms like x^2.
The two axes are independent. You can have soft and symmetric, which is a smooth tanh waveshaper with mainly third and fifth-order odd harmonics, very analog but tonally cool. You can have soft and asymmetric, which is the typical tube-style plugin with a strong second harmonic that reads as fullness. Hard and symmetric is a digital hard clipper, screaming odd harmonics. Hard and asymmetric is the territory of broken transistors and some fuzz pedals.
The tube myth
The widely repeated rule "tubes produce warm even harmonics, solid state produces harsh odd harmonics" is an oversimplification that falls apart on inspection.
Most tube guitar amps with EL34 or 6L6 power stages, and most hi-fi tube power amps, use push-pull output topology. A push-pull stage drives two output devices 180 degrees out of phase, then combines them through an output transformer. The signal components add. The even-order distortion components, which appear in phase at both devices, subtract and cancel in the load. The result is a stage that, by topology, produces predominantly odd harmonics, the same family as a hard clipper. This is documented in Wikipedia's Tube Sound entry, Aiken Amps' technical writing, and Robert Till's harmonic distortion notes.
Where does the famous "tube warmth" actually come from. Two places. First, single-ended Class A tube stages, which are inherently asymmetric and do produce strong even harmonics, particularly the second. Second, frequency-dependent transformer behaviour, where the core saturates more easily at low frequencies than at high ones. Neither of those is universal to "tubes" as a category. Many tube circuits, especially push-pull power sections, share their harmonic family with the very transistor circuits they are supposed to sound different from.
Tape is the next victim of the same loose framing. The idealised magnetic hysteresis curve is symmetric, so a pure tape transfer function produces primarily odd harmonics, not even. Real tape decks introduce some asymmetry through bias adjustment, so even harmonics do show up, but they are not the dominant component. What people perceive as tape warmth is largely the time-domain hysteresis behaviour, where the output depends on the recent history of the magnetic state, rather than even-order harmonic content.
How the two axes guide a mixing decision
Once the axes are separated, plugin choice becomes a series of straightforward questions.
For bus glue and mix warmth, a soft, asymmetric saturator with frequency-dependent behaviour is the right tool. Tape and transformer emulations sit here. The asymmetry gives a noticeable second harmonic, an exact octave above the fundamental, that reads as fullness. The dynamic envelope softens transients enough to feel like gentle compression without a compressor in the chain.
For drum-bus character that needs to keep the attack, a soft symmetric waveshaper works well. Clean tanh-style behaviour with no hysteresis. It adds odd-harmonic edge that helps drums cut through without rounding the initial transient.
For surgical peak control, a hard clipper with oversampling enabled is the correct choice. Symmetric hard clipping gives a fast, predictable peak cap. Placed before a limiter in the mastering chain, it absorbs the impulsive content that would otherwise force the limiter to gain-reduce the whole signal. The limiter then operates on more uniform program material with less audible pumping. This clipper-then-limiter order is documented in Mastering.com's chain notes and is standard practice in modern loudness work.
For Aphex-style excitement, a high-passed branch run through an asymmetric saturator, then mixed back in parallel, adds perceived brightness without boosting the existing high end. The added even harmonics sit in octave relationships with the source, which is what makes the result feel musical rather than just bright.
The aliasing footnote
Any nonlinearity creates harmonics that exceed Nyquist and fold back as aliasing. A 5 kHz sine passed through a hard clipper at 44.1 kHz produces a third harmonic at 15 kHz (fine), a fifth at 25 kHz (folds back to around 19 kHz), and a seventh at 35 kHz (folds back to around 9 kHz, where it sits inside the program material as an inharmonic artefact).
The standard fix is oversampling the nonlinearity, typically 4x or 8x, which is why most modern saturators and clippers offer an oversampling switch. Bandlimited methods using continuous-time convolution and integrated bandlimited ramps are newer alternatives explored in the DAFx and EUSIPCO literature, but oversampling remains the default in commercial plugins.
One related practical detail. A signal that does not clip at the sample points can still clip during D/A reconstruction if the analog waveform between samples exceeds 0 dBFS. True-peak limiters use roughly 4x oversampling to detect this. The common ceiling for streaming masters is -1.0 dBTP.
The practical takeaway
Choose a saturator for harmonics combined with dynamic envelope behaviour. Choose a clipper for fast peak reduction with no envelope behaviour. Choose something labelled "distortion" when the listener is meant to hear the effect as an effect. The marketing label on the box is a hint at intent, not a description of mechanism. The mechanism lives on the two axes, and those are what you are actually picking.