This entry is part of the Chart Index, the reference library for the Chart Design Field Guide.

Parallel coordinates is the form that lets you visualise many continuous variables at once. Each variable becomes a vertical axis; each observation becomes a polyline crossing every axis at its value. Where line patterns cluster, you find sub-populations; where lines cross between two axes, you find inverse relationships; where lines bundle, you find correlated variables.

The form was introduced by Alfred Inselberg in the 1980s as a way to visualise high-dimensional data in a 2D plane. It is dense, abstract, and uniquely powerful for multivariate exploration. It is also the chart most often misused — one of the easiest forms to make beautiful and impossible to read at the same time.

What it is

A parallel coordinates plot maps N continuous variables to N parallel vertical axes, equally spaced along the x-axis. Each observation is rendered as a polyline that visits each axis at the value of that observation's corresponding variable. With many observations, the polylines form a dense fabric — the shape of which reveals relationships, clusters, and outliers across all dimensions simultaneously.

The hero chart above plots twenty-four car models across six attributes, coloured by body class. Read it left to right: price, power, engine size and kerb weight rise together, so the lines run in three parallel bands — hatches low, SUVs in the middle, utes on top — and stay in their lanes. Between weight and fuel use the bands hold; between fuel use and acceleration they cross, because the heavy utes are both the thirstiest and the slowest. Randomise the attributes and the fabric shivers but the three bands stay put: the structure is stronger than the noise.

When to use it

Parallel coordinates is the right choice when:

  • You have 3 or more continuous variables to compare across observations.
  • The reader's question is "what relationships exist among these variables, and are there sub-populations?"
  • You have many observations (50–5,000) — the fabric of overlapping lines is the form's power.
  • The audience is willing to learn the form or already trained.
  • You want multivariate pattern discovery rather than precise reading.

When not to use it

  • One or two variables. Use a scatter plot.
  • Discrete categories. Parallel coordinates needs continuous values per axis. For mixed types, use coordinated multiple views.
  • Audiences unfamiliar with the form. The reading task is non-obvious. Reserve for analytical contexts or expect to teach the form.
  • Precise value reading. The form shows patterns; specific numbers require tooltips.

Design principles

Order axes deliberately

The axes' arrangement determines what relationships are visible. A pair of adjacent axes is, in effect, a scatter plot turned inside out: correlated variables produce parallel segments, inversely related variables produce a fan of crossings, and unrelated variables produce a tangle. Non-adjacent pairs show nothing at all. The two figures below plot the same twenty-four cars with the axes in two orders.

Axes ordered by the story: the related variables sit side by side
Price, power, engine and weight are adjacent, so the lines run in three parallel bands and the classes are obvious. The fan of crossings between fuel use and acceleration is the one inverse relationship in the set, placed where the eye will find it.
Axes in alphabetical order: the same cars, the structure scrambled
Identical data with the axes sorted by name. Acceleration now sits beside engine size, so the first gap is a fan of crossings; power sits beside price, which is fine, but weight is banished to the far right where nothing relates to it. Every pattern is still in the data. Half of them are no longer in the picture.

Use opacity for many lines

A parallel coordinates plot with hundreds of fully opaque lines is a wall. Reduce the opacity so dense regions become visible as darker fabric without losing the individual lines. The two figures below plot the same 120 sensor readings, first solid and then at 15 percent.

120 readings at full opacity: a solid wall
Five variables from 120 sensor readings, every line solid. Somewhere in there are two populations of readings, but the later lines paint over the earlier ones and the picture is a single blue sheet.
The same readings at 15 percent opacity: two populations appear
Nothing changed but the alpha. Where lines pile up the fabric darkens, and the hot, dry, windy, dusty days separate from the cool, humid, still ones as two distinct bands on every axis. Translucency turns overplotting into density.

Highlight categories or sub-groups

When you know there are sub-populations — car classes, customer segments, weather regimes — colour the lines by group. The eye picks out the bundling of each group across the axes; one group may bunch high on one axis and low on another, telling a story no single bivariate plot would. The hero chart does this with three classes. Take the colour away and the classes are still in the data, but the reader has to discover them.

The hero's cars in one colour: the classes have to be inferred
Same twenty-four models, no group encoding. The three bands are still visible on the left-hand axes because the classes are so well separated, but at fuel use and acceleration the lines interleave and there is no way to tell a heavy SUV from a light ute.

Normalise axes thoughtfully

Each axis can be scaled to its own range (relative) or to a common scale (absolute). Relative scaling reveals patterns within each variable's range; absolute scaling shows how variables differ in magnitude. Most parallel coordinates use per-axis min–max — every figure on this page does, with the range pinned in the axis ticks — and the choice should be stated, because a line that runs flat across two axes means "same rank on each", not "same value".

Add small histograms beside each axis

A small histogram or density curve beside each axis shows the marginal distribution of that variable. The reader gets both the multivariate pattern (the polylines) and the per-variable distribution (the histograms).

Annotate notable lines or clusters

If the chart's story is here is the cluster of efficient cars or here is the one outlier model, highlight those lines or groups with colour and a label. The dense fabric otherwise has no narrative direction.

Avoid overwhelming the form

The form's complexity tempts addition: 12 axes, 5,000 lines, three colour encodings, animated transitions. Most of those additions degrade legibility. Be ruthless about what dimensions and observations the chart actually needs.

Anatomy

The composition of a parallel coordinates plot
An anatomical guide

A parallel coordinates plot's anatomy is a row of vertical axes, polylines crossing them, variable names above, ticks beside, and an optional colour encoding for group. The form's information density is high; its design needs to remain quiet around the polylines.

  • Scatter plot matrix (SPLOM) — all pairwise bivariate scatters in a grid. The conventional alternative for multivariate exploration.
  • Radar / spider chart — radial parallel coordinates. Different reading task.
  • Heatmap (small multiples) — for the same N-variable comparison without crossings.
  • Andrews curves — function-based visualisation of high-dimensional data; statistical cousin.
  • Chord / arc diagram — for relationships among categorical variables.

Reading list

  • Inselberg, A. (1985). The Plane with Parallel Coordinates. The original paper.
  • Inselberg, A. (2009). Parallel Coordinates: Visual Multidimensional Geometry and Its Applications. The definitive reference.
  • Heinrich, J. & Weiskopf, D. (2013). State of the Art of Parallel Coordinates. A survey of variants and developments.
  • Munzner, T. (2014). Visualization Analysis and Design. On axis ordering and the limits of the form.