How to Make a Violin Plot in Python (Seaborn & Matplotlib Guide)

A violin plot shows the full distribution of a variable across groups by mirroring a kernel density estimate around a central axis. Where a box plot collapses the data to five summary numbers, a violin reveals shape — skew, multiple peaks, and where the data actually concentrates. That makes it the right choice when distribution shape carries the scientific message. This guide builds one in seaborn (the fast path) and matplotlib (the precise path), and covers the options that matter for real figures.

New to the idea? Start with what is a violin plot, then come back here for the code.

1. A basic violin plot in seaborn

Seaborn plots violins directly from a tidy DataFrame. Pass the grouping column to both x and hue with legend=False to color each group without a redundant legend. Notice how group B's two bulges expose a bimodal distribution that a box plot would have flattened.

import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np

rng = np.random.default_rng(42)
df = pd.DataFrame({
    "score": np.concatenate([
        rng.normal(70, 8, 120),
        np.concatenate([rng.normal(60, 4, 60), rng.normal(82, 4, 60)]),  # bimodal
        rng.normal(75, 12, 120),
    ]),
    "group": np.repeat(["A", "B", "C"], 120),
})

fig, ax = plt.subplots(figsize=(6.2, 4.6))
sns.violinplot(data=df, x="group", y="score", hue="group",
               palette=["#6b7280", "#7c3aed", "#ec4899"], legend=False, ax=ax)
ax.set(xlabel="", ylabel="Score")
sns.despine()
plt.tight_layout()

2. Control the inside of the violin

The inner argument decides what is drawn inside each violin: "box" for a mini box plot, "quart" for quartile lines, "point" or "stick" for raw observations, or None for a clean shape. Pair it with cut=0 so the density does not imply values beyond your data range.

import seaborn as sns
import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(6.2, 4.6))
sns.violinplot(data=df, x="group", y="score", hue="group", legend=False,
               inner="box",          # mini box plot inside each violin
               cut=0,                 # do not extend density past the data range
               palette="mako", ax=ax)
ax.set(xlabel="", ylabel="Score")
sns.despine()
plt.tight_layout()

3. Overlay the raw data points

For small samples, a smooth density can over-promise. Set inner=None and overlay a strip plot so reviewers see the actual observations behind the shape. This is the most defensible form of a violin plot for experimental data.

import seaborn as sns
import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(6.2, 4.6))
sns.violinplot(data=df, x="group", y="score", hue="group", legend=False,
               inner=None, palette="Greys", cut=0, ax=ax)
sns.stripplot(data=df, x="group", y="score", color="#7c3aed",
              size=2.5, jitter=0.12, alpha=0.5, ax=ax)
ax.set(xlabel="", ylabel="Score")
sns.despine()
plt.tight_layout()

4. Split violins for two-group comparison

When every category has exactly two subgroups (placebo vs treatment, male vs female), a split violin puts one distribution on each half of the same shape. This doubles the information density without cluttering the axis. Use split=True with a two-level hue.

import seaborn as sns
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np

rng = np.random.default_rng(7)
n = 150
df2 = pd.DataFrame({
    "value": np.concatenate([
        rng.normal(50, 8, n), rng.normal(54, 8, n),   # site 1
        rng.normal(60, 8, n), rng.normal(72, 8, n),   # site 2
    ]),
    "site": np.repeat(["Site 1", "Site 2"], 2 * n),
    "arm": np.tile(np.repeat(["Placebo", "Treatment"], n), 2),
})

fig, ax = plt.subplots(figsize=(6.6, 4.6))
sns.violinplot(data=df2, x="site", y="value", hue="arm", split=True,
               gap=0.1, inner="quart", palette=["#6b7280", "#7c3aed"], ax=ax)
ax.set(xlabel="", ylabel="Response")
ax.legend(title="", frameon=False)
sns.despine()
plt.tight_layout()

5. Pure matplotlib (no seaborn)

If you cannot add seaborn as a dependency, matplotlib has its own ax.violinplot(). It takes a list of arrays and returns the artist collections, so you style the bodies yourself. Use showmedians=True to mark the center.

import matplotlib.pyplot as plt
import numpy as np

rng = np.random.default_rng(42)
data = [rng.normal(loc, 8, 200) for loc in (70, 66, 78)]
labels = ["A", "B", "C"]
colors = ["#6b7280", "#7c3aed", "#ec4899"]

fig, ax = plt.subplots(figsize=(6.2, 4.6))
parts = ax.violinplot(data, showmedians=True)

for body, color in zip(parts["bodies"], colors):
    body.set_facecolor(color)
    body.set_alpha(0.65)

ax.set_xticks(range(1, len(labels) + 1))
ax.set_xticklabels(labels)
ax.set_ylabel("Score")
ax.spines[["top", "right"]].set_visible(False)
plt.tight_layout()

Violin plot vs box plot: which to use

Common mistakes and fixes

• Density extends past real data — set cut=0 so the violin stops at the observed min and max.
• Violin on tiny samples — KDE invents smooth shape from few points. Below ~20 per group, overlay raw points or switch to a box plot.
• Over-smoothed or spiky shape — tune bw_adjust (smaller is bumpier, larger is smoother) to match the data.
• Redundant legend — when coloring by the x variable, set hue to the same column and pass legend=False.
• No explanation of width — state in the caption that width encodes estimated density, since not all readers know the convention.

Frequently asked questions